Escolar Documentos
Profissional Documentos
Cultura Documentos
MODDE
By Umetrics
Version 9
© 1992-2009 MKS Umetrics AB, all rights reserved
Information in this document is subject to change without notice and does not
represent a commitment on part of Umetrics. The software, which includes
information contained in any databases, described in this document is
furnished under license agreement or nondisclosure agreement and may be
used or copied only in accordance with the terms of the agreement. It is
against the law to copy the software except as specifically allowed in the
license or nondisclosure agreement. No part of this user guide may be
reproduced or transmitted in any form or by any means, electronic or
mechanical, including photocopying and recording, for any purpose, without
the express written permission of Umetrics.
Umetrics patents and trade marks:
OPLS/O2PLS®, SWE-9802229-6, USA-6754543,
OSC®, SWE-0000563-7, USA-6853923
PLS-TREE®, Pending
SEQUENTIAL MODELING, Pending
DESIGN SPACE, Pending
MODDE®, RED - MUP®, VALUE FROM DATA®, OPLS®, O2PLS®, O2-
PLS®, OPLS-DA®, O2PLS-DA™
PLS-TREE™, S-PLOT™, EZinfo®, SBOL®, FABSTAT®, BATCH
FINGERPRINT®, SIMCA®.
ID # 2038
MKS Umetrics AB
Stortorget 21
SE-211 34 Malmö
Sweden
Phone: +46 (0)40 664 2580
Email: info@umetrics.com
Welcome
Welcome to the user guide for MODDE 9. This is your guide to MODDE and
its capabilities.
Assumed prior knowledge
We assume that you are already familiar with Windows. You should be
familiar with the topics discussed in your Microsoft Windows User's Guide,
including how to use the mouse select, click, shift-click, control-click, press,
drag, and choose from a menu. You should also know how to search through
directories to find files.
Content
This user guide is divided into 21 chapters. Chapter 1 gives a short
introduction of how to use MODDE. Chapter 2 presents an introduction to
experimental design. Chapter 3 is an overview of the program. Chapters 4 - 20
provide step-by-step procedures for creating and using experimental designs
with MODDE. Chapter 21 describes how the Help function works.
Appendix A, "Statistical notes" presents short explanations of statistical
methods used by MODDE.
Appendix B, “Designs” presents short descriptions of the designs available in
MODDE.
Appendix C: "Optimizer" describes the optimizer feature and the properties of
the different optimizer objectives.
Appendix D: "Design Space" describes the design space estimation feature.
References are available on the references page.
User guide edition Tuesday, September 22, 2009.
Table of Contents
How to get started with MODDE 1
Installation ..................................................................................................................... 1
Starting MODDE ........................................................................................................... 1
Experimental cycle ........................................................................................................ 1
Design phase .................................................................................................................. 2
Defining factors................................................................................................................... 2
Defining responses .............................................................................................................. 2
Defining objective ............................................................................................................... 2
Analysis phase ............................................................................................................... 2
Explore the data (Worksheet menu) .................................................................................... 2
Evaluate the design ............................................................................................................. 3
Fit ........................................................................................................................................ 3
Review the fit using plots and lists ...................................................................................... 3
Diagnostics .......................................................................................................................... 3
Interpret the model .............................................................................................................. 3
Refine the model ................................................................................................................. 4
Prediction phase (using the model) ................................................................................ 4
Introduction to MODDE and experimental design 5
General description ........................................................................................................ 5
What is modeling and experimental design? ................................................................. 5
Objectives of modeling and experimental design .......................................................... 5
Screening models and designs ....................................................................................... 5
Number of factors in screening designs .............................................................................. 6
Number of factors with split objective ................................................................................ 6
Response surface modeling (RSM) designs................................................................... 6
Number of factors in RSM designs ..................................................................................... 6
Number of factors in Split objective ................................................................................... 7
Fit methods .................................................................................................................... 7
Multiple Linear Regression (MLR) ..................................................................................... 7
Partial Least Squares (PLS)................................................................................................. 7
Results ................................................................................................................................. 8
Analysis phase ............................................................................................................... 9
Review the model fit ........................................................................................................... 9
Assess model adequacy ....................................................................................................... 9
Prediction - using the fitted model ................................................................................. 9
Convention................................................................................................................... 10
Limitations in investigation names.................................................................................... 10
Limitations in factor and response names ......................................................................... 10
Case sensitivity ................................................................................................................. 10
Menu reference syntax ...................................................................................................... 10
Suggestions for further reading on experimental designs ............................................ 10
Overview 11
Application icon and symbol ....................................................................................... 11
i
User Guide to MODDE
ii
Table of Contents
Add factor, add response, add experiment, insert rows ............................................... 32
Sort............................................................................................................................... 32
Sorting the candidate set ................................................................................................... 33
Model / reference mixture............................................................................................ 34
Model ................................................................................................................................ 34
Reference mixture ............................................................................................................. 36
Generators .................................................................................................................... 37
Edit the model and/or generators before creating the worksheet ................................. 38
View 39
Introduction.................................................................................................................. 39
Toolbars ....................................................................................................................... 39
Standard toolbar ................................................................................................................ 39
Spreadsheet toolbar ........................................................................................................... 41
Plot toolbar ........................................................................................................................ 41
Model toolbar .................................................................................................................... 44
Window toolbar................................................................................................................. 44
Status bar........................................................................................................................... 44
Dockable windows....................................................................................................... 44
Analysis advisor ................................................................................................................ 45
Audit trail .......................................................................................................................... 45
Favorites............................................................................................................................ 45
Output ............................................................................................................................... 50
Notes ................................................................................................................................. 50
Add to Favorites .......................................................................................................... 51
Add to Report .............................................................................................................. 51
Full Screen ................................................................................................................... 51
Changing the default options using Investigation Options .......................................... 52
Plot labels .......................................................................................................................... 52
Number format .................................................................................................................. 52
Audit trail .......................................................................................................................... 53
Alpha level ........................................................................................................................ 54
Coefficients ....................................................................................................................... 55
Blocking ............................................................................................................................ 56
Confidence level ............................................................................................................... 56
List presentation ................................................................................................................ 57
R2 in plots ......................................................................................................................... 57
Residuals ........................................................................................................................... 58
Select Factor...................................................................................................................... 58
General Options ........................................................................................................... 58
General page ..................................................................................................................... 59
List Options ....................................................................................................................... 60
Restore .............................................................................................................................. 61
Customize .................................................................................................................... 61
Customize toolbars ............................................................................................................ 61
Customize commands ....................................................................................................... 61
Customize options ............................................................................................................. 62
Customize options ............................................................................................................. 62
Design 63
Introduction.................................................................................................................. 63
iii
User Guide to MODDE
Responses 75
Introduction.................................................................................................................. 75
Response definition dialog........................................................................................... 75
Response name .................................................................................................................. 76
Abbreviation...................................................................................................................... 76
Units .................................................................................................................................. 76
Selecting type of response ................................................................................................. 76
Limits ................................................................................................................................ 76
Regular responses ........................................................................................................ 77
Transformation .................................................................................................................. 77
MLR scaling...................................................................................................................... 77
PLS scaling ....................................................................................................................... 77
Derived responses ........................................................................................................ 78
Defining derived responses ............................................................................................... 79
Modifying a derived response ........................................................................................... 80
Copying or deleting a derived response ............................................................................ 80
Using sets of variables in derived responses ..................................................................... 80
Syntax for derived responses............................................................................................. 81
Operators and functions in derived responses ................................................................... 81
Qualitative factors in derived responses ............................................................................ 82
Linked responses ......................................................................................................... 82
Response definition spreadsheet .................................................................................. 83
Printing the response definition spreadsheet ..................................................................... 83
Response manipulations in short ................................................................................. 84
Opening the responses spreadsheet ................................................................................... 84
iv
Table of Contents
D-Optimal 103
What are D-Optimal designs? .................................................................................... 103
When do I use D-Optimal designs? ........................................................................... 103
D-Optimal pages in the design wizard ....................................................................... 104
Design generation criteria section ................................................................................... 104
Design alternatives section .............................................................................................. 106
Candidate set section ....................................................................................................... 107
v
User Guide to MODDE
Analysis 133
Introduction................................................................................................................ 133
Organization of the Analysis menu............................................................................ 133
Evaluate ..................................................................................................................... 134
Condition number ........................................................................................................... 134
Runs, terms and degrees of freedom ............................................................................... 134
Analysis Wizard......................................................................................................... 135
Open the Analysis Wizard............................................................................................... 135
Analysis Wizard content ................................................................................................. 136
vi
Table of Contents
Prediction 165
Introduction................................................................................................................ 165
Contour plot wizard ................................................................................................... 166
Inner plot type ................................................................................................................. 166
Selecting responses ......................................................................................................... 166
Using constraints ............................................................................................................. 166
Plot options ..................................................................................................................... 167
2D contour ...................................................................................................................... 167
4D contour ...................................................................................................................... 169
Response Surface ............................................................................................................ 172
Prediction plot wizard ................................................................................................ 173
Prediction plot wizard first page selections ..................................................................... 173
Axes and constants .......................................................................................................... 174
Overlay prediction plot ................................................................................................... 174
Response prediction plot ............................................................................................ 175
Factor type ...................................................................................................................... 175
Prediction list ............................................................................................................. 176
Scatter plot ................................................................................................................. 176
Sweet spot plot ........................................................................................................... 177
Creating a sweet spot plot ............................................................................................... 177
Optimizer ................................................................................................................... 180
Factor spreadsheet in the Optimizer ................................................................................ 181
vii
User Guide to MODDE
viii
Table of Contents
ix
User Guide to MODDE
x
Table of Contents
xi
User Guide to MODDE
References 289
Index 291
xii
How to get started with MODDE
Installation
You can install MODDE under Windows Vista, XP, and 2000.
Starting MODDE
Start MODDE by double-clicking its icon.
• To read about the MODDE software look in Help (contains the same
information as the user guide).
• To run tutorial examples, find them at www.umetrics.com (Downloads),
select an example, open the investigation used in the tutorial (.mip-file
included with the installation in the Investigation folder) and follow the
analysis steps.
• To start a new investigation, on the File menu click New.
Experimental cycle
The experimental cycle consists of three phases:
1. The design phase where you define your factors and within which ranges
they should be varied, your responses, objective, design and model.
2. The analysis phase where you explore your data, review the raw data and the
fit, review diagnostics in plots and lists, refine and interpret the model.
1
User Guide to MODDE
3. The prediction phase where you use the model to predict the optimum area
for operability.
Design phase
Once you have clicked New on the File menu the design wizard is then opened. Enter
the name and location of the new investigation in the appropriate fields.
For details on the design wizard, see the Design Wizard chapter.
Defining factors
On the second page of the design wizard, enter factor names, ranges, etc.
For details on defining factors see the Factors chapter.
Defining responses
On the third page of the design wizard, enter response name(s), etc.
For details on defining responses see the Responses chapter.
Defining objective
On the next page of the design wizard, click your objective Screening, RSM, or Split
Objective. Clicking Paste Data disables Next; clicking Finish opens the worksheet
for pasting.
Given your objective, the type of factors (formulation or process) and the number of
factors, MODDE recommends a design, a model, and a number of center points. The
total required number of runs for this design is displayed. Other choices compatible
with your objective type and number of factors are listed with the required number of
runs.
After clicking Finish, the Worksheet is automatically generated for all classical
designs. With D-Optimal designs, clicking Next opens a dialog for the selection and
generation of the D-Optimal design and worksheet.
After making the experiments, fill in the response values and actual factor values in the
worksheet.
For details on the objective pages, see the Objective, model and design chapter. For
details on the worksheet, see the section Worksheet spreadsheet in the Worksheet
chapter.
Analysis phase
After the response values have been entered in the worksheet you can review the raw
data, fit the model, review the fitted model, interpret the model, and refine the model.
2
How to get started with MODDE
Fit
When you are ready to fit a model to your design you click Fit on the Analysis menu.
MODDE automatically fits using MLR when the condition number is low and there are
no mixture factors. The fit methods available are MLR, PLS and for mixtures several
variants.
For PLS models you can choose to extract more components than the MODDE default.
To extract more components click Next Component on the Analysis menu. Next
Component is unavailable when MLR is used as fit method.
Diagnostics
MODDE has a number of diagnostic plots:
• Residual plots to find outliers, drifts, trends etc.
• Box Cox Plot to select the best transformation of Y.
• ANOVA: ANalysis Of VAriance, in particular review the Lack of Fit. The
estimation of lack of fit is only available when there are replicated points as
it compares the pure error and the model error.
3
User Guide to MODDE
When PLS is used for regression, scores and loadings can be plotted. These plots
provide an overview of the data. On the Analysis menu, click PLS Plots to select the
score or loading plot you want to display.
To remove outliers or insignificant model terms use the interactive exclude tool .
Click the button and then click/mark the outlier/term the in a plot. You can also
exclude it in the worksheet, by right-clicking the specific cell and clicking Exclude
value(s). The model is automatically refitted.
Note: When excluding an outlier or model term in a plot, the outlier or model
terms is only excluded for the displayed response.
Remove/Add terms from/to the model that are insignificant/significant for all
responses. Use Edit | Model or the interactive tool with the coefficient plot.
After refining your model you should once more review the fit and diagnostics as
described above.
Note: You can have a different model for each response in the same
investigation.
4
Introduction to MODDE and
experimental design
General description
MODDE - (MODeling and DEsign) is a Windows program for the generation and
evaluation of statistical experimental designs.
Methods of statistical experimental designs have evolved since the pioneering work of
Fisher in 1926. These methods, further refined by Box, Hunter, Scheffé, Tagushi and
others, provides the users with a powerful methodology for efficient experimentation.
The application icon for MODDE is a pink circle with a gray M inside.
5
User Guide to MODDE
In the screening stage one uses simple models (linear or linear with interactions), and
experimental designs that allows the identification of the factors with the largest effects
in the fewest possible number of experimental runs.
MODDE supports: Full Factorial, Fractional factorial, L-designs, Plackett Burman,
Rechtschaffner, Onion, and D-optimal designs for screening experiments.
With mixture factors, MODDE supports the classical axial design when the region is a
simplex.
6
Introduction to MODDE and experimental design
Fit methods
The data collected by the experimental design is used to estimate the coefficients of the
model. The model represents the relationship between the response Y and the factors
X1, X2, etc.
MODDE uses multiple linear regression (MLR) or Partial Least Squares (PLS) to
estimate the coefficients of the terms in the model. MODDE recommends PLS when
the investigation has a high condition number.
7
User Guide to MODDE
Results
Both MLR and PLS computes regression coefficients for each response. Thus Y is
expressed as a function of the X's according to the selected model (i.e. linear, linear
plus interactions, or quadratic).
8
Introduction to MODDE and experimental design
Analysis phase
All results of model fitting, by MLR or PLS are displayed, in the same way,
graphically and in lists.
9
User Guide to MODDE
Convention
Limitations in investigation names
The following characters cannot be used: = \ / : * “ ? < > |.
The length (including path) cannot be larger than 260 characters.
Case sensitivity
MODDE is case insensitive. Lower or upper case in names will be displayed as
entered, but for all comparisons lower or upper case are considered the same.
10
Overview
The symbol for MODDE is a gray cube with a pink M (above). The application icon is
a pink circle with a grey M inside (below).
Investigation
Experimental plans in MODDE are organized into investigations. You can think of an
investigation as a file folder containing all of the information related to a particular
experiment. When you select or open a given investigation you can access, display and
use all of its information. This information is organized in the following components:
factors, responses, constraints, inclusions, candidate set, model, design, worksheet,
analysis and predictions.
Managing investigations
Investigations are binary files saved by MODDE with the extension *.mip. When you
start a new investigation you select its name and destination directory.
You can open, save, save as, and delete investigations.
You can generate the fold over of investigations with fractional factorial designs or
Plackett Burman.
You can double-click a MODDE investigation (a *.mip file), in Microsoft Explorer, to
open that investigation.
11
User Guide to MODDE
MODDE does not save the fitted model. In order to review the results of the analysis
and use make predictions, you need to fit the model by clicking Analysis | Fit or
having Automatic fit turned on (default) in View | General Options tab General
when the investigation is opened. After the model has been fitted, you can use the
Analysis menu to create plots and list to review the model and fit and use the
Prediction menu to create prediction plots and lists.
Organization
The MODDE window consists of a command menu bar and toolbars.
12
Overview
The menu bar consists of the menus File, Edit, View, Design, Worksheet, Analysis,
Prediction, Show, Window, and Help.
File menu
Under the File menu the following is available:
• New: Start a new investigation.
• Open: Open a MODDE investigation.
• Close: Close a MODDE investigation.
• Save: Saves the current investigation.
• Save As: Saves the current investigation under a specified path and name.
• Delete Investigation: Deletes the current investigation.
• Save Plot/List A : Saves the active plot or list.
• Revert: Reverts to the last saved copy of the investigation.
• Send by email: Sends the current Investigation by e-mail.
• Generate HTML report: Automatic report generator in HTML format.
• Complement Design: Select to complement your design by:
1. Fold over: Complements screening designs of resolution III or IV
by adding the fold over.
2. Estimate square terms in a screening design: complements the
design to support a quadratic model for a selection of factors.
3. Complement the design with D-Optimal to support a specified
model.
4. Screening to RSM Rechtschaffner: Complements the
Rechtschaffner screening design to the RSM Rechtschaffner.
5. Complement the Doehlert design.
6. Plackett Burman Super-Saturated to Plackett Burman:
Complements PBSS to the regular PB.
• Protect Investigation: Allows for a password protected investigation and
locking of investigation.
13
User Guide to MODDE
Edit menu
From the Edit menu you can:
• Undo
• Cut/Copy/Paste and Delete
• Select All
• Row insertion, addition
• Sort
• Edit the Model
• Edit the Reference Mixture
• Edit the Generators of fractional factorial designs
View menu
The following is available from the View menu:
• Toolbars: View/Hide Toolbars.
• Dockable windows: Show or hide dockable windows.
• Add to Favorites: Add the current plot or list to the Favorites dockable
window.
• Add to Report: Add the current plot or list to the current report generated by
MODDE.
• Full screen: Maximize the plot area.
• Investigation Options: Change the default plot and list options.
• General Options: Change the default options.
• Customize: Customize the toolbars.
Design menu
Use the Design menu for the following:
• Factors and Responses: Define and modify factors and responses
• Constraints and Inclusions: Define constraints and inclusions
• Objective: Specify and modify objective, model and design
• D-Optimal: Generate and evaluate D-Optimal designs, open the candidate
set for viewing or editing, evaluate onion plots.
• Design wizard: Open the design wizard
14
Overview
Worksheet menu
Use the Worksheet menu to:
• Edit the Worksheet.
• Set Run Order.
• Open the Curvature Diagnostic Plot.
• Create the Scatter plot.
• Open the Histogram plot.
• Open the Descriptive statistics plot and list.
• Open the Correlation plot or matrix.
• Open the Replicate plot.
Analysis menu
Use the Analysis menu to:
• Evaluate the design
• Analyze each response in the Analysis Wizard
• Select Fit Method (MLR, PLS etc)
• Fit model by MLR or PLS
• Extract the Next Component: Add a PLS component
Review the fit
Results displayed as tables, lists, or graphs
• Summary of the fit
• PLS Summary
• PLS Plots
• Coefficients and Effects
• VIP Plot (PLS only)
Investigate diagnostics
• Residuals Plots
• Box-Cox Plot (MLR only)
• ANOVA
Prediction menu
With the fitted model you can click the Prediction menu and:
• Display contour plot from the Contour plot wizard including mixture
contour plots and response surface plots (3D).
• Create prediction plots.
• Make predictions in the Prediction spreadsheet.
15
User Guide to MODDE
Show menu
Use the Show menu to display the following: objective, design matrix, design
region, confounding, model, D-Optimal summary, and reference mixture.
Window menu
Use the Windows menu to access the standard window commands.
Toolbars
Click Toolbars from the View menu to display or hide the following:
1. The Standard toolbar consist of command buttons for immediate action and
menu commands.
2. The Spreadsheet toolbar displays all spreadsheets.
3. The Plot toolbar is for zooming in and out on plots, reading plot values and
coordinates, and drawing regression lines.
4. The Model toolbar (model information) summarizes the investigation. The
summary consists of the number of factors and responses, the number of
runs, the Objective, Design and model and the Fit method.
5. The Window toolbar displays buttons for arranging the open windows and
Full Screen.
6. The Status bar displays an explanation to the button that the mouse points to.
These toolbars are described in detail in the View chapter.
16
File
Introduction
Use the File menu to create, open, close, save, and delete investigations, save active
plot or list, revert to last saved investigation, send the current investigation by e-mail,
generate a HTML report, complement the design, lock or password protect the
investigation, for print purposes and to close MODDE.
New
To create a new investigation click File | New.
This first page allows you to:
• Enter the name of the new investigation.
• Select the location where to save.
• Select the design type: Traditional Designs or Advanced Designs including
RED-MUP and D-Optimal designs from an imported candidate set or scores
from a SIMCA-P model.
• Select to import a design from file.
• Enable the audit trail by selecting the Enable the Audit trail check box.
17
User Guide to MODDE
Click Next to open the design wizard. The design wizard guides you through the
design phase, from factor and response definition, objective, design and model
selection, up to the generation of the worksheet.
For a detail description of the design wizard see the Design Wizard chapter.
Traditional designs
When you create a new design Traditional Designs is by default selected. The
traditional designs include classical designs such as factorial, Plackett Burman, L-
designs, Rechtschaffner, and also D-Optimal and onion designs. For details, see the
Design appendix.
Advanced designs
When you select Advanced Designs and click Next the Advanced Designs box is
opened with the available designs:
• RED-MUP.
• Onion or D-Optimal designs from imported candidate set.
• Multivariate onion or D-Optimal designs from scores.
18
File
RED-MUP
RED-MUPs are designs available for 96 (8x12), 384 (16x24), and 1536 (32x48) runs.
The designs are built from sub-designs.
To create a RED-MUP:
1. In the first page of the design wizard click Advanced Designs and then
Next.
2. Click RED-MUP and then Next.
3. Define all factors for the two sub-designs and then click Next.
4. Define the responses and then click Next.
5. Select the objective for both the vertical and the horizontal designs:
Screening or RSM.
6. Leave the factors that should be included in the vertical design with fewer
runs to the left.
7. Move the factors that should be included in the horizontal design to the right.
8. In this page you can also select the number of plates used, and if applicable
the plate factor(s) that contain plate information.
9. Select the desired plate size in the Plate-Size box and optionally select the
Plate/Block factor interactions check box.
10. Click Next to select the vertical design then click Next again to select the
horizontal design. MODDE adds center points when the selected design does
not fill up the plate size.
Note: Some special RED-MUP designs, that aim to aim to make better use of
the plate, are available for the 96 well plates (8 x 12).
19
User Guide to MODDE
Note: Only regular factors are imported here. If your candidate set contains
qualitative or formulation factors you have to enter the factors and settings in
the Factor Definition and import the candidate set from the D-Optimal page.
The designs available here are Onion and D-Optimal designs. Onion is only available
when there are enough experiments in comparison with the number of factors. For
more on Onion designs, see the D-Optimal Onion designs section in the Design
appendix and the Tutorial.
Multivariate designs
To create a design using the scores from a SIMCA-P project as factors:
1. On the Advanced Designs page click Multivariate onion or D-Optimal
designs from scores.
2. Click the Browse button under Import scores (factors) from a .usp created
in SIMCA-P 9 or higher to select the SIMCA-P project.
3. SIMCA-P opens automatically. Return to MODDE without closing SIMCA-
P.
4. Select the model from the Model box and click Next to import the factors
from SIMCA-P. The scores are then automatically loaded from the SIMCA-
P .usp-file, and the candidate set for the onion design comprised of all
objects (rows) in the workset of the SIMCA-P model is selected as the basis
of the onion design.
20
File
Open
To open a MODDE investigation, click Open on the File menu. In the Open-dialog
select the directory and you get a list of all the MODDE Investigation Projects (*.mip)
present in that directory. Double-click the desired investigation to open it or mark it
and click Open.
To open investigations created in MODDE version 4 and earlier, switch to Old
MODDE Files (*. ini) in Files of type to view them, mark the .ini and click Open to
convert it to the current format.
Delete investigation
To delete the current investigation click Delete Investigation on the File menu. It is
not possible to delete any other investigations than the current from MODDE.
For read-only investigations Delete Investigation is unavailable.
Save plot/list as
Plots can be saved as Enhanced Meta Files (*.emf), Bitmap files (*.bmp), Jpeg files
(*.jpg) or PNG files (*.png). Lists can be saved as Text files (*.txt) or Web Pages
(*.htm, html).
21
User Guide to MODDE
To save a plot, display it, make sure it is active and then click File | Save Plot As. The
Save Format dialog is displayed.
In the dialog check the Keep aspect ratio box to keep the proportions of the plot as on
the screen or according to a predefined plot size.
In the Plot size box you can choose between Custom, 640x480(VGA),
800x600(SVGA), 1024x768, Original size.
The predefined formats are useful primarily when displaying MODDE plots on a small
screen.
To save a list, display it, make sure it is active and then click File | Save List As. The
Save dialog is then displayed.
In the Save as type box, select file type .txt or .htm/html.
Revert
To revert to the investigation as when it was last saved, click Revert on the File menu.
Send by E-Mail
To send an investigation by e-mail click Send by E-Mail on the File menu. Note that
the investigation sent is the current investigation as it was last saved.
22
File
A placeholder tells MODDE the desired item to fill from the current investigation
when you click Update Report. If you add plots and lists as placeholders and save the
template, you can generate a report in the desired format, for any investigation, by
selecting the saved template and clicking Update Report.
For details, see the Report Generator chapter.
Complement design
To complement a design is to make a new investigation consisting of the design, or a
part of the design, of the current investigation plus its complement. See also the section
Inclusions vs. complement design in the chapter Constraints and Inclusions.
Use Complement Design when you want to:
• Estimate separately a set of terms (interactions, or main terms and
interactions) that were confounded in a Resolution III or IV fractional
factorial design.
• Complement a screening design to an RSM design supporting the full
quadratic model.
• Complement a screening design to estimate selective curvature effects.
• Add additional experimental runs to improve the quality (i.e. the condition
number or G-efficiency) of an existing set of experiments.
• Use already performed experiments in a Doehlert design to set up a new
Doehlert moving the center of the design or add a factor. For more on the
Doehlert designs see the Design appendix.
• Use already performed experiments in a Super-Saturated Plackett Burman
design to add experiments resulting in a regular Plackett Burman design.
To complement the current design, click Complement Design on the File menu and
the Complement Design Wizard is opened. Click the desired complementing method
and click Next.
23
User Guide to MODDE
Fold over
When you choose to complement your design with fold over MODDE makes a new
investigation consisting of the design of the current investigation plus its fold over
(complement). The fold over design has as many experimental runs as the original
design.
Fold over designs are available for fractional factorial design of resolution III or IV
and Plackett Burman designs.
With the complete design (original + fold over), all main effects are clear from 2
factors interactions. With resolution III and IV designs MODDE automatically adds a
block factor. You may remove the block factor from the model in the menu Edit |
Model.
To fold over your design:
1. Click File | Complement Design.
2. Click Fold over and then click Next.
3. Enter the name and location of the new investigation. It is recommended to
add an additional center point to detect a shift in the mean.
4. Click Finish and the new investigation opens.
24
File
D-Optimal
Complementing a design d-optimally is the most flexible way of complementing a
design.
25
User Guide to MODDE
Mixture factors
Complementing the design d-optimally with mixture factors, after giving the name and
location of the new investigation, the D-Optimal wizard opens. Select the new model.
The displayed number of runs includes the original design runs. Click Next, the wizard
guides you in generating the new investigation.
The displayed number of runs includes the original design runs as inclusions. Click
Next; the wizard guides you in generating the new investigation.
26
File
Complement Doehlert
Doehlert designs can be complemented by expanding the design region or adding a
factor.
To complement a Doehlert design, click Doehlert in the Complement Design Wizard
and click Next.
27
User Guide to MODDE
Select how to complement, click Next, enter the name, location of the new
investigation and number of additional center points, and then click Finish to generate
the new investigation.
28
File
Protect investigation
When you click File | Protect Investigation you can select to Encrypt/Password
protect or Lock the investigation.
Lock investigation
Click File | Protect Investigation | Lock to lock the investigation. Locked
investigations are automatically fitted when opened. Any plot or list can be displayed,
but you cannot make any changes in the investigation. The investigation becomes
'Read only' with the exception of the prediction spreadsheet. If you click Permanently,
unlock on Save As, a copy of the investigation can be unlocked by using File | Save
as. When Permanently is selected the investigation cannot be unlocked.
It is possible to also encrypt and password protect the investigation by selecting the
Encrypt / Password Protect investigation check box. The investigation is then
encrypted. This check box is optional; select it only if you want the investigation to be
password protected.
29
User Guide to MODDE
Print Setup
Change the page orientation, printer, and printer settings by clicking Print Setup on
the File menu.
Plots are preferably printed with Orientation Landscape.
The print menu is only valid for an active list or plot.
Print Preview
On File, click Print Preview and the current window can be view as it will be printed.
Print Preview is only available when there is an open plot or list.
Print
Print is only available for the active plot or list. With the plot or list active, use one of
the following manners to print:
• On File menu click Print.
• Right-click the plot or list and click Print.
Exit
Click Exit from the File menu to close the MODDE program.
30
Edit
Introduction
Use the Edit menu to undo, cut, copy, paste, delete, select all, add experiment/insert
rows/add factor/add response, sort, edit model, edit reference mixture, and edit the
generators.
Undo
Click Edit | Undo or press CTRL+Z to undo changes in MODDE. Such changes can
be changing values or text (factor name for instance), cutting, copying, pasting,
deleting, sorting in spreadsheets, or editing the model.
Undo is activated after changes in:
• The spreadsheets Worksheet, Factors, Responses, Constraints, Inclusions,
Prediction, and Optimizer.
• The model terms in the Model Properties dialog (Edit | Model) or by
clicking the Exclude-button and marking model terms to exclude in a plot.
• Plots displaying experiments (observations) by clicking the Exclude-button
and marking model terms to exclude in a plot.
Undo remembers the ten last actions in the worksheet, in the edit model dialog, and in
plots displaying experiments (observations) or model terms.
Undo works on the active plot or spreadsheet.
Note: After editing the values in worksheet the undo feature of plots
displaying experiments stops working as manual changes in the worksheet
empty the undo-memory for such plots.
31
User Guide to MODDE
Sort
Sorting is available for the worksheet, constraints, inclusions, candidate set, and
predictions list.
To sort a list, right-click and click Sort or click Edit | Sort.
In the dialog, select the column to sort the list by from the Select the column to sort
box and click the Add Column button. The column appears in the list with the default
sort type. Click the sort order you want under Sort selected, Ascending or
Descending.
If a column of the spreadsheet is marked when sort is activated it becomes, by default,
the primary column to sort by.
Add more columns to select secondary and tertiary etc., columns to sort by. Specify for
each column the sort order. Use the Remove button to remove a column from the sort
list or drag the item outside the list.
The sorting starts when OK is clicked.
32
Edit
33
User Guide to MODDE
Model
Click Edit | Model or the Edit Model button to modify the model. The Model
Properties dialog opens on the Model Terms page and you can edit the model by
adding or deleting terms.
Use the For Response box to specify for which response you want to alter the design,
if not for all.
A quadratic term is represented by factor1*factor1 and a cross term (interaction) by
factor1*factor2. A cubic term is represented by factor1*factor1*factor1 and a three
factor interaction is represented by factor1*factor2*factor3.
34
Edit
Add terms
To add terms in the Model Properties dialog use one of the following methods:
1. Mark the factors to the left and click the relevant buttons Factors,
Interactions, Squares, or Int. & Sq. (interactions and squares)
2. Double-click a factor or mark it and click the left arrow =>; note that it is
added in the New term field. Continue until the new term is fully defined
and click the second arrow => adding the term to the right side.
3. Position the cursor in the New term field and type the abbreviation of the
factors to include using ‘*’ to separate them.
4. To add three factor interactions, mark the three factors, click the left arrow
and then on the right =>. Or type the abbreviations in the New term field.
Remove terms
To remove terms in the Model Properties dialog, mark the term(s) and click the
Remove button in the dialog.
Undo
The Undo button becomes active after making a change in the model. Clicking Undo
only reverses the last change in the Model Properties dialog.
Reset
Click the Reset button to reset the model to the default model.
35
User Guide to MODDE
Reference mixture
In Model Properties, click the Reference Mixture tab to switch to the reference
mixture page.
Use this dialog to change the default reference mixture of a Cox model. The reference
mixture is used to impose constraints on the coefficients of the Cox mixture model.
The default reference mixture is the centroid of the experimental region. If the region is
a regular simplex with q mixture factors, the centroid is the point with coordinates 1/ q,
1/ q,....1/ q. If the experimental region is irregular, the centroid is the constrained
centroid of the irregular constrained region.
Click the Reset button to restore the reference mixture to the default values.
36
Edit
Generators
Click Generators on the Edit menu to open the Generators dialog.
A generator is a column of signs in the extended design table of the basic factors that is
used to introduce additional factors in the fractional factorial designs.
For example, let us assume that 5 factors are to be investigated in 8 runs. The extended
design table is the table of the full factorial in three factors (basic factors), symbolically
named a, b and c plus the additional columns for all the interactions. Any
interaction column can be used to introduce additional factors. Let us say that to
introduce the 2 additional factors, d, and e, the column of signs of a*b and a*c are
selected. Then d = ab and e = ac are the generators of the fractional factorial design 25-2
(see Box, Hunter and Hunter for further information).
When MODDE generates fractional factorial designs the default generators used are
those published in Box, Hunter and Hunter.
Editing and/or changing the default generators of a design is done in order to estimate
selected interactions in a fractional factorial design of resolution III or IV instead of the
default.
MODDE supports the choice of positive or negative generators.
To edit a generator, click in the generator column of the desired row, and enter a new
generator. The confounding, in the confounding column, is updated.
When you click OK, your design and worksheet are deleted and new ones are
generated.
37
User Guide to MODDE
38
View
Introduction
Using the View menu you can show/hide the toolbars, show/hide the dockable
windows Analysis Advisor, Audit trail, Favorites, Output, and Notes, and execute the
commands: Add to Favorites, Add to Report, make window Full screen, change the
default investigation options, change the general options, and customize toolbars,
commands and options.
Toolbars
On the View menu, click Toolbars to view the available toolbars. Click the name of
the toolbar to show or hide it. All toolbars can be floating or docked beneath the menu
bar.
Clicking a button on a toolbar will perform a certain task. As a guide to what happens
when a button is clicked, a short text will appear when you hold the pointer over a
button. Text will also appear in the Status bar at the bottom of the screen.
You can customize all toolbars according to your wishes. For more, see Customize
later in this chapter.
Standard toolbar
The Standard toolbar consists of a number of command buttons and the response box.
The buttons listed below are available.
39
User Guide to MODDE
Click the Objective button to open the design wizard. If none or too few factors
are defined the design wizard opens on the factor page, otherwise it opens on the
objective page where you select the objective of the design.
Click the Worksheet button to open the worksheet for editing or viewing.
Edit model
Click the Edit model button to edit the model or reference mixture when
applicable. Model Terms is the default page opened. To edit the reference mixture,
click the Reference Mixture tab.
Response box
Use the Response box to change responses in the active plot or list.
40
View
The last response check box cannot be cleared. To get single selection when switching
between responses using the response combo box, hold down the CTRL-button for
single selection functionality.
Spreadsheet toolbar
The Spreadsheet toolbar contains buttons for all MODDE spreadsheets. In the order of
the toolbar: the factor, response, worksheet, prediction, and optimizer spreadsheets.
Plot toolbar
The Plot toolbar holds buttons that enable you to select type of marking to be used
when marking areas in plots, insert labels or text in plots, zoom in and out, read
positions in graphs, do linear regression in scatter plots, get information about
observations (experiments) or variables (model terms or responses), or exclude
experiments and model terms.
The tools are activated when you click them.
41
User Guide to MODDE
Arrow
The arrow button is used to select which type of marking to use.
Click the small arrow to the right of the arrow and click the type of marker from the
menu.
• Free Mark: Allows marking to take any shape.
• Block Mark: Marks in a rectangular shape.
• X Mark: Vertical bar.
• Y Mark: Horizontal bar.
• No Mark: The arrow works as a data reader displaying the number, name,
and value of plotted data points when hovering near the data point symbol.
Text
Click the T-button to insert a text-field in a plot. The field can contain text or a
value and be inserted in all plots but the contour, response surface, sweet spot, and 3D
scatter plots.
Open the plot, click the T-button, click where you want to position the text-field, type
the text or numbers, and then press ENTER or click somewhere else in the plot.
MODDE will continue to open a new text-field each time you click until the T-button
is pressed again. This label can be removed by clicking it again and emptying the cell.
Zoom in
Click the arrow near the Zoom in button , and select the type of zoom from the
drop down menu:
• Scale xy: Magnifies a rectangular region
• Scale x: Expands the x direction
• Scale y: Expands the y direction
• Subplot: Magnifies a subplot in a multiplot.
And then mark the desired region of the plot to zoom.
Zooming in a scatter 3D plot is described in the Zoom and rotate section in the Plots
and lists chapter.
Zoom out
Click the Zoom out button to revert zoom to original scale in the steps taken
when zooming.
Coordinate Reader
The coordinate reader displays the coordinate values of any point of the plot area.
Click the Read coordinate values button and mark the point of interest and the
coordinate reader will display the coordinates at that point.
42
View
Regression line
The regression line and equation can be displayed for any 2D scatter plot in MODDE.
Open a scatter plot, for instance the Observed vs. predicted plot, and click the
Rotate
For rotating a scatter 3D or response surface plot see the Zoom and rotate section in
the Plots and lists chapter.
43
User Guide to MODDE
Model toolbar
The Model toolbar displays the summary of the current investigation.
1. Factors: The total number of factors (the number of uncontrolled).
2. Responses: The number of responses.
3. Runs: The number of runs included in the design (not counting rows set as
'Excl' in the worksheet).
4. Objective: The objective, Screening, RSM or Split Objective.
5. Design: The selected design and model.
6. Method of Fit: MLR, PLS etc.
Window toolbar
The Window toolbar commands are general Windows commands and work
accordingly. Click the Window toolbar commands Full screen (press the F9 key to get
back to regular), Cascade Windows, Tile Windows horizontally or Tile Windows
vertically to arrange the windows.
Status bar
The Status bar displays the explanation to the command button that the mouse points
to.
Dockable windows
MODDE has five dockable windows: analysis advisor, audit trail, favorites, output,
and notes.
Open the dockable windows by clicking Dockable Windows on the View menu and
then clicking the desired dockable window.
When you click the Auto Hide button the window will hide away. If you then
hover with the mouse over the dockable window name, it slides out and becomes
visible. If you want your window to stay visible after it has slid out, just click the pin
button again .
44
View
Analysis advisor
The Analysis Advisor is automatically activated after fitting an investigation. The
advisor explains the analysis plots and results.
Show or hide the Analysis Advisor window is available by clicking View | Dockable
Audit trail
When turned on, the audit trail logs changes done to the investigation.
To open or close the audit trail, click Dockable Windows on the View menu, and then
click Audit Trail.
For more, see the Audit trail subsection in the Changing the default options using
Investigation Options section.
Favorites
To open or close the favorites window use one of the following methods:
45
User Guide to MODDE
Adding favorites
To add an open plot or list to favorites, right-click it and click Add to Favorites or
select it from the View menu. Note that this adds the plot with the selected settings to
the favorites.
You can also right-click in the Favorites window and select Add Command to
Favorites.
Favorites window
Right-clicking the Favorites window opens the menu below. A description of the
menu items follows.
Adding favorites
To add an open plot or list to favorites, right-click it and click Add to Favorites or
select it from the View menu. Note that this adds the plot with the selected settings to
the favorites.
You can also right-click in the Favorites window and select Add Command to
Favorites.
46
View
Opening all items in the folder displays the coefficient plot and observed vs. predicted
plot for the first response tiled.
Click it to open all the items. This gives the same result as Open All Items in Folder
but with a single click.
47
User Guide to MODDE
Rename
All folders, commands, plots, spreadsheets, and lists can be renamed according to your
wishes by right-clicking it and clicking Rename.
Delete
All items in Favorites can be deleted by right-clicking the item and clicking Delete, or
marking it and pressing the DELETE-key on your keyboard.
Add command to favorites
A number of commands can be added to favorites. The commands from the menu and
toolbars are displayed on the left, under Categories, and the items available under that
command are displayed to the right.
For example, if you click Prediction under Categories, all the commands on the
Prediction menu are displayed to the right.
Click a command, for example Prediction Plot Wizard, and click Add. The command
is added to the Favorites window. You may continue adding commands, and when
done, click Close.
48
View
When a command has a submenu, double-click it to view the submenu items. For
example Analysis | Residuals, double-click Residuals and the submenu of four items
it displayed. Select the desired one and click the Add button.
49
User Guide to MODDE
Note: When importing a favorites file, the new favorites will replace the
current favorites. If you want to keep you current configuration and switch
back to it later, export to file before importing.
To save the current favorites configuration to file:
1. Right-click the Favorites window.
2. Click Export.
3. Enter the name and location in the Save As dialog,
4. Click Save.
To import favorites from .xml-file:
1. Right-click the Favorites window.
2. Click Import.
3. Browse for the file in the Open dialog.
4. Click Open.
Restoring favorites
To restore Favorites to the MODDE default, click General Options on the View
menu, click the Restore tab and click the Restore button after Restore favorites.
Output
The output window is a log book of the session. All MODDE messages and actions are
recorded in the Logbook.
Open the output window by clicking Dockable windows on the View menu and
clicking Output.
To make this window smaller, pull the top and it will remember its size. Double-click
the window caption to make it floating or make it dock.
Hint: While dragging the window hold down the CTRL key. It won't dock if
you are close to the frame of the main frame window.
Notes
In Notes you can record your own notes concerning the investigation. You can
paste MODDE plots and lists, and embed objects such as Microsoft Excel
spreadsheet, word, etc. All these commands are available by right-clicking the window.
This file can then be saved as .rft (Rich Text Format) and read directly by a word
processor with all plots.
Open the notes window by clicking Dockable windows on the View menu and
clicking Notes.
50
View
Add to Favorites
Plots and lists can be added to the Favorites dockable window. For more, see the
Favorites section earlier in this chapter.
Add to Report
Plots and lists can be added to the HTML report generator. For more, see the Report
Generator chapter.
Full Screen
You can use the Full screen command to maximize the plot area. Full screen can be
toggled on and off using F9 and is available from the Window toolbar as well as from
the View menu. When Full Screen is selected from the menu, the Window toolbar is
automatically opened as a floating toolbar.
51
User Guide to MODDE
Plot labels
Select the plot label to display as default on plots displaying the experiments, for
example residual plots, by clicking None, Experiment Number, Experiment Name,
or Run Order taken from the worksheet.
Number format
Click the number of decimals 2 - 5 or the scientific (E) format in lists by clicking
Auto, .00, .000, .0000, .00000, or Scientific.
When exporting unscaled coefficients to use in other applications, use the scientific
format to get maximum precision in the coefficients.
52
View
Audit trail
When the Audit Trail is turned on each investigation in MODDE has a separate audit
trail. Each audit trail consists of one or more sessions that in turn consist of events. A
new session is started and appended to the audit trail when an investigation is opened,
and ends when the investigation is saved.
In addition to logging events, MODDE logs information about the user, and date and
time of the events.
To view the audit trail, click the Audit Trail tab in the Output / Notes / Audit trail
dockable window. If this tab is not shown, display it by clicking View | Dockable
Windows | Audit Trail.
53
User Guide to MODDE
Alpha level
Select the significance level by clicking 1%, 5%, or 10% to be used in the Lack of
fit (LoF) plot. The default is 5%.
54
View
Coefficients
In MODDE you can select the type of coefficients you want to display. The default is
Scaled and centered coefficients.
PLS orthogonal
The PLS orthogonal coefficients are available only when fitting the model with PLS.
The PLS orthogonal coefficients are the PLS model coefficients re-expressed to
correspond to factors centered and scaled using orthogonal scaling (coded as -1 to 1).
PLS orthogonal coefficients are not available in investigations with formulation factors
only.
With process and mixture factors, the PLS orthogonal coefficients refer to process
factors scaled orthogonally, and mixture factors unscaled (original units).
Note: The PLS orthogonal coefficients are meant for comparison with
MLR only. They are incorrect when the design is not balanced and the mean is
not equal to the mid-range.
Normalized
To make the coefficients comparable when responses (Y's) have different ranges, you
can select to display the coefficients in Normalized form that is, the coefficients are
divided by the standard deviation of their respective response. The normalized mode is
the system default for the Coefficient Overview Plot.
55
User Guide to MODDE
Unscaled
The unscaled coefficients are the coefficients corresponding to unscaled, uncentered
data. When exporting unscaled coefficients to use in other applications, be sure to use
the E-format to get maximum precision in the coefficients.
Blocking
You can select to have the block factor treated as a fixed or random effect and the
predictions computed accordingly.
Select the block factor as Fixed effect when the external variability can be set at will
and the primary objective for blocking is to eliminate that source of variability
Select the block factor as Random effect when the external variability cannot be
controlled and set at will, and the primary objective is to make prediction without
specifying the block level, and taking into account the external variability.
See the Statistical Appendix for more information.
Confidence level
Select the confidence level 90%, 95%, or 99% for computation of the confidence
interval on the model coefficients, effects, and predictions. The system default is 95%.
56
View
List presentation
By default lists are displayed using the Regular - shows all orthogonal settings
option, when there are qualitative factors. And the design matrix by default displays
The current Worksheet scaled and centered.
R2 in plots
Click R2 or R2 adjusted to display in the Summary plot and PLS Summary plot. The
default is to display R2.
57
User Guide to MODDE
Residuals
Select the type of residuals to be used in the residual plots.
Raw residuals
The raw residual is the difference between the observed and the fitted (predicted)
value.
Standardized residuals
The standardized residual is the raw residual divided by the residual standard deviation
(RSD).
Default
With MLR and 2 or more degrees of freedom, deleted studentized residuals are the
MODDE default when plotting residuals.
With PLS and models with less than 2 degrees of freedom, MODDE uses as default the
Standardized residuals.
Select Factor
Select the factor to display as default in the Residuals | vs. Variable plot and Effects |
Main Effect plot.
General Options
The General Options has three pages with settings that are used for all investigations.
• General with general settings such as automatic fit.
• List Options like coloring and grid styles.
• Restore with buttons to restore plot settings, favorites, and ‘don’t show
again’-messages.
Click General Options on the View menu to access the options.
Click the Show Details button to display a short description of the marked option.
58
View
General page
The General page is divided into the parts General, Investigation Settings, Program
Limits, and Audit Trail.
General
Under General you can change the following options:
• With Automatic fit = Yes, each investigation opened is automatically fitted.
This is MODDE's default. If you set Automatic Fit = No MODDE will not
fit when you open, nor when you make changes in the worksheet, but will
still fit when you make changes using the interactive exclude tool.
• With Automatically display log = Yes, the Output window is automatically
opened when opening an investigation. This is MODDE's default.
• In Most recent file list you can enter the number of investigations you want
to see in the recent file list under the menu File. MODDE's default is 6.
• When Show expanded design factors in RED-MUP worksheet = No, the
expanded design factors will be hidden. This is MODDE's default. For more
see the section on RED-MUP in the Design appendix.
• In Theme you can change the look of MODDE. The default is Office 2003
Classic Theme. Available are also Office 2000, Office XP Classic Theme,
Windows XP, and Whidbey.
• OpenGL plots = No is the default and recommended for the Contour and
Sweet Spot Plots. In 3D scatter plots, OpenGL is always used.
59
User Guide to MODDE
Investigation settings
The available investigation settings are:
• Max number of layers in Onion Design to limit how many layers that will
be possible to make. MODDE's default is 10.
• Replicate tolerance is the number used when MODDE decides whether
experiments can be considered to be replicates or not. MODDE's default is
0.1, that is, 10% of the range for each factor.
• Show units in worksheet = Yes results displaying the specified factor and
response units in parenthesis after the factor resp. response names.
Program limits
Under Program Limits you can set the Maximum candidate set size.
The size of the candidate set in MODDE is by default limited to 512 000 rows when
MODDE creates the candidate set for you. You can change this limit.
The maximum size of the candidate set that you can create and generate a design from
is limited by the RAM in your computer.
Audit trail
Prompt for extended user info, found under Audit Trail, is by default No. If you
want MODDE to prompt the user to enter extended user information, change this
setting to Yes.
List Options
Use this page to change the default colors on lists. Other options available in tab list
options are:
• Default number format
• Print lists in color
• Headers in bold
• Threshold for the correlation matrix
60
View
Restore
Use the Restore page to restore Umetrics defaults for Plot Settings, Favorites, and
'don’t show again'-messages.
Customize
Use Customize to customize toolbars and menus as in Office 2003 and later.
Click Customize on the View menu to open the dialog.
With the Customize dialog open you can interactively customize all available toolbars
and menus by dragging an item to a new location.
Customize toolbars
On the Toolbars page you can:
• Select which toolbars to display by clearing or selecting the available
toolbars.
• Create new toolbars by clicking New.
• Rename toolbars you have created
• Delete toolbars you have created.
• Reset MODDE toolbars to their default content and position.
Customize commands
On the Commands page you can customize available menus and create new menus.
61
User Guide to MODDE
Customize options
On the Options page you can select:
• To Always show full menus.
• To display Large icons.
• To Show ScreenTips on toolbars.
• To Show shortcut keys in ScreenTips.
• Menu animations among: (System default), Random, Unfold, Slide,
Fade, None.
Customize keyboard
On the Keyboard page you can select to assign and reassign shortcut keys to new and
old commands.
62
Design
Introduction
When you create or modify a design, some or all items on the Design menu are used.
The available items are:
• Factors and Responses: Define and modify factors and responses.
• Constraints and Inclusions: Define constraints and inclusions.
• Objective: Specify and modify objective, model and design
• D-Optimal: Generate and evaluate D-optimal designs, open the candidate set
for viewing or editing, view onion plots.
• Design wizard: Open the design wizard.
Factors
Click Factors on the Design menu to open the factor spreadsheet.
To modify one of the factors listed, mark it and press the ENTER key on your
keyboard or right-click the factor and click Edit.
To add a new factor, double-click the last line or right-click the spreadsheet and click
Add Factor.
For details, see the Factors chapter.
Responses
Click Responses on the Design menu to open the response spreadsheet.
To modify one of the responses listed, mark it and press the ENTER key on your
keyboard or right-click the response and click Edit.
To add a new response, double-click the last line or right-click the spreadsheet and
click Add Response.
For details, see the Responses chapter.
Constraints
Click Constraints on the Design menu to open the Constraints dialog.
To add a constraint, use the spreadsheet part or the graphical part of the dialog.
For details, see the Constraints section in the Constraints and inclusions chapter.
63
User Guide to MODDE
Inclusions
Click Inclusions on the Design menu to open the inclusions dialog.
In this dialog you can do the following:
• Import runs from a text file or another investigation to inclusions.
• Import the current worksheet.
• Add the current inclusion runs to the current worksheet.
• Open the D-Optimal page in the design wizard.
For details, see the Inclusions section in the Constraints and inclusions chapter.
Objective
Click Objective on the Design menu to open the objective pages.
In the objective pages you define the purpose of the design by selecting Screening,
Response Surface Modeling (RSM), Split Objective, or Paste Data. Click Next to
select which design you want MODDE to create. When selecting Paste Data, clicking
Finish opens the worksheet allowing you to paste the data.
For details, see the Objective, model and design chapter.
D-Optimal
When you have created a D-Optimal design a number of items are available under D-
Optimal on the Design menu. For all D-Optimal designs you can click:
• Generate to regenerate the design.
• Candidate set to view the current candidate set and which design runs that
are included in the current design.
• Evaluate to view the D-Optimal results and select another one of the
generated designs.
• Design Plot to create the D-Optimal design plot.
For Onion designs you can additionally create onion plots by clicking Onion plot and
Onion 3D Scatter. These plots display the current candidate set colored according to
layer and the experiments currently included colored black.
For details, see the D-Optimal chapter.
Design wizard
The design wizard is automatically opened when starting a new investigation from File
| New. It can also be opened after creating an investigation by clicking Design Wizard
on the Design menu or by pressing CTRL+W.
The design wizard includes pages for creating/modifying factors, creating/modifying
responses, defining/modifying a constraint numerically, objective and design selection,
and when applicable the D-Optimal pages with the possibility to define inclusions.
For details, see the Design Wizard chapter.
64
Factors
Introduction
Factors are variables that can be varied, or vary, during an experiment. A typical
example of a factor is amount of raw material or temperature.
Open the Factor spreadsheet by clicking Design | Factors or File | New, and then
Next.
In the Factors spreadsheet (window), you define (enter), modify, and delete factors.
MODDE supports quantitative, qualitative, and mixture factors.
Quantitative factors may be used in a transformed metric. When factors are
transformed, the design is made in the transformed units, but the worksheet is
expressed in original units.
65
User Guide to MODDE
The Factor Definition dialog is divided in an upper part and a lower part. The upper
part displays Factor name, Abbreviation, and Units and is available independently of
what is displayed in the lower part of the dialog. The lower part has two pages:
General, which is the default page when opening the dialog, and Advanced.
Factor name
Enter the Factor name with up to 50 alphanumeric characters.
Abbreviation
The Abbreviation is automatically filled with the first 3 characters of the factor name.
You can change the abbreviation as desired using up to 5 characters.
The abbreviation is used as plot label in plots, in the Model and Confoundings lists
found under the Show menu, in Worksheet | Correlation | Matrix, and in Edit |
Generators.
Units
Enter the unit of the factor (optional). The units are displayed in the factor spreadsheet
and can optionally be displayed in the worksheet, see the General page section in the
View chapter for more.
General page
On the General page of the factor definition dialog you select which type of factor you
are defining in Type of factor, the factor settings in Low and High or Settings, and
how it is used in Use.
66
Factors
Note: MODDE is limited in the precision of the factor values. Factors ranging
over a larger range than that, for instance. low at 0.0001 and high at 10000,
cannot be treated as a factor.
For quantitative multilevel and qualitative factors the levels planned to be used should
be entered. For a qualitative factor each entry in Settings has to be text, optionally
including numbers, while for quantitative multilevel each entry has to be a number.
Note: In the same experiment you can have both mixture factors and regular
process factors defined as quantitative or qualitative. Up to 12 factors are
allowed when both process and formulation factors are defined in the design.
Quantitative (default)
Quantitative factors are continuous factors defined at two levels, Low and High. To
define more than two levels, see Quantitative multilevel.
Up to 32 factors are allowed for screening designs and up to 20 for RSM designs.
Quantitative multilevel
To specify more than two levels for a quantitative factor click Quantitative multilevel
under Type of Factor.
MODDE supports up to 24 levels for quantitative multilevel factors. Constraints are
not allowed with this type of factor, and the available designs are D-Optimal, Mixed
full factorial, and three level designs when applicable.
Up to 32 factors are allowed for screening designs and up to 20 for RSM designs.
Qualitative
To specify a qualitative factor, click Qualitative under Type of Factor.
Qualitative factors are discrete. For a qualitative factor, the levels should not stand in
relation to each other. If the levels are a range although discrete then the factors should
be defined as quantitative multilevel.
MODDE supports up to 24 levels for qualitative factors.
RSM designs cannot be created with only qualitative terms. With one or more
quantitative terms present up to 20 extended qualitative terms are allowed for RSM
designs.
67
User Guide to MODDE
Formulation
To specify a mixture factor, click Formulation under Type of Factor.
Define a mixture factor as Formulation, when it is not an inert filler. Define the
experimental range of the mixture formulation factor by entering its Low and High
values.
Up to 20 formulation factors are allowed in screening designs and up to 16 in RSM
designs.
Filler
Specify a mixture factor as filler, by clicking Filler under Type of Factor, when:
• It is always present in the mixture. That is, the sum of the High of the other
mixture factors does not exceed 1.
• It accounts for a large percentage of the mixture.
• There are no restrictions on its range. Rather that factor is added at the end to
bring the mixture total to the desired amount, 1 (100%).
• You are NOT interested in estimating the effect of the filler per se.
For a filler factor, the experimental range Low and High values are grayed out as it
will be calculated as 1 minus the sum of the other mixture factors.
Use
Under Use you can select how the factor will be used in the design. A factor can be
Controlled, Uncontrolled, or Constant.
Controlled (default)
When you can control the settings of the factor, leave the default under Use
(Controlled).
These factors can be regular process factors (i.e. pH, Temp, etc.) defined as
quantitative, quantitative multilevel or qualitative, or mixture factors, defined as
formulation or filler.
A Filler factor can only be defined as Controlled. Uncontrolled and Constant are
unavailable for a filler factor.
68
Factors
Uncontrolled
Define a factor as Uncontrolled (under Use) if you cannot control it but want to
measure and record its value. Examples of such factors are ambient temperature or
humidity.
Mixture factors (defined as Formulation or Filler) cannot be uncontrolled and the
option Uncontrolled is therefore unavailable.
Constant
Define a factor as Constant (under Use) when you want the worksheet to display a
setting of a factor that is not changed.
Quantitative, qualitative, and formulation factors can be defined as Constant factors.
Quantitative and qualitative constant factors are displayed in the worksheet and
counted in the model toolbar.
When mixture factors are constant, the mixture total T for the controlled mixture
factors is equal to: T = 1 - Σ[constant mixture factors].
Multilevel quantitative and filler factors cannot be defined as Constant.
Advanced
For quantitative and quantitative multilevel factors the Advanced tab offers the
possibility to transform and change the MLR scaling. For regular quantitative factors it
is also possible to define number of decimals to be used for the factor from this page.
69
User Guide to MODDE
Transformation
To transform a factor, click the Advanced tab, and in the Transform box click the
transformation of your choice.
When you transform a factor, the design is created in the transformed units, but the
worksheet is expressed in original units. Hence transformation of a factor will change
the center point and the star point values in the worksheet.
All transformed factors are displayed with a “~” (tilde) near the name in lists and plots.
The following transformations are available:
Transformation Description
None Default
Lin C1 * Y + C2
(C1 * Y + C2)C3
Power
where C3 can be any value from -2 to 2.
When a transformation is selected (except None), the relevant constant fields are
displayed.
The field C3 is only displayed for the power transformation.
70
Factors
Mixture factors
When fitting the model with MLR, the mixture factors are not scaled.
You can select to fit the model in pseudo components. This is recommended when the
mixture region is regular.
For investigations containing both process and mixture factors, by default process
factors are orthogonally scaled and the mixture factors are unscaled. The coefficients
displayed as scaled and centered correspond to this default scaling of the variables i.e.
mixture unscaled and process orthogonally scaled. If you select to display the unscaled
coefficients, they correspond to all factors unscaled, including the process factors.
Note: You should select the same scaling for all the factors, the system default
is recommended.
PLS scaling
When fitting the model with PLS, all factors including mixture factors are always
scaled and centered to unit variance.
For mixture factors, when you select pseudo components, the mixture factors are first
transformed to pseudo components and then scaled to unit variance (pseudo
components can be switched on/off).
Number of decimals
In the Advanced page of the factor definition dialog you can select Number of
decimals.
The values for number of decimals are: Free, 0, 1, 2, 3, 4 and represent the number of
digits displayed after the decimal point.
The Number of decimals-value should correspond to the precision with which the
factor can be set in your equipment. It is important that it is not set too low since after
setting this value all values for that factor will be rounded accordingly in the
worksheet. If you do not know the precision of the instrument, leave No. of decimals
'Free'. 'Free' means that no rounding off of the results and values corresponding to this
factor will take place.
For example, the settings of a factor in a CCC design, is its high value (in
orthogonal scaled units) multiplied by the 4th root of the number of runs in the factorial
part of the design. If the precision of this factor is set to 0 all decimal digits are
removed from this factor setting in the worksheet and computation will take place
using the values in the worksheet.
71
User Guide to MODDE
Factors can be added by double-clicking the last row of the factor definition
spreadsheet or right-clicking the spreadsheet and clicking Add Factor.
To copy factors, mark the factors to copy, click Edit | Copy or press CTRL+C, and
click Edit | Paste or press CTRL+V. MODDE copies the factors and adds a digit after
the name when pasting to make it unique.
72
Factors
Adding a factor
To add a factor through the factor definition dialog use one of the following methods:
• Double-click the last line of the factor definition spreadsheet.
• Right-click the spreadsheet and click Add Factor.
• Open View | Design Wizard and click New in the Define factors page.
Quantitative factors can be added by typing on the last (empty) row starting with
entering the factor name, see the example:
Modifying a factor
It is possible to edit the fields Name, Abbr., Units, and Settings directly in the
spreadsheet.
To modify any of the other fields, double-click that factor in the factors spreadsheet.
The Factor Definition dialog opens with the attributes of the factor to modify.
The Factor Definition dialog can also be opened by clicking View | Design
Wizard and clicking Edit in the Define factors page.
Copying a factor
Factors can be copied and pasted in the factor spreadsheet by clicking Edit | Copy and
then Edit | Paste or pressing CTRL+C and then CTRL+V.
73
User Guide to MODDE
Deleting a factor
To delete a factor, mark it in the factor definition spreadsheet and use one of the
following methods:
• Press the DELETE key on the keyboard.
• Right-click and click Delete.
• Click Edit | Delete.
You are warned of the possible consequences. Click Yes/No in the confirmation box.
74
Responses
Introduction
A response is the result from an experiment. A typical example of a response is yield.
Open the response definition spreadsheet from Design | Responses or File | New,
Next, Next.
In the response definition spreadsheet, you define (enter), modify, delete, copy, print,
and list responses. MODDE supports only quantitative responses.
Responses may be transformed, and MODDE supports several transformations. For
transformed responses, predictions, contour plots and 3D plots, are back transformed to
original units.
75
User Guide to MODDE
Response name
Enter the response name with up to 50 alphanumeric characters in the Response name
box.
Abbreviation
The Abbreviation is automatically filled with the first 3 characters of the response
name. You can change the abbreviation as desired using up to 5 characters.
The abbreviation is used as plot label in plots and in Worksheet | Correlation |
Matrix.
Units
Optionally enter the unit of the response in the Units box. The units are displayed in
the response definition spreadsheet and can optionally be displayed in the worksheet,
see the General page subsection in the View chapter for more.
Limits
Fill in the Min, Target, and Max fields when that information is available to you.
These values are then automatically used in the Design Space, Sweet Spot, and
Optimizer windows.
76
Responses
Regular responses
Regular responses are the standard responses measured and fitted in the current
investigation. Regular responses can be transformed and it is also possible to change
the modifier for PLS scaling.
Transformation
The following transformations are available:
Transformation Description
None Default
Lin C1 * Y + C2
(C1 * Y + C2)C3
Power
where C3 can be any value from -2 to 2.
When a transformation is selected (except None), the constants in the formula are
entered in the fields displayed after selecting a transformation. The C3 field is only
displayed for the power transformation.
Specifying a transformation for a response is done to get the best mathematical fit of
the estimated function.
MLR scaling
When fitting the model with MLR no scaling of responses is available.
PLS scaling
When fitting the model with PLS it is possible to scale to unit variance with or without
a modifier.
With the default scaling option, the responses are centered and scaled to unit variance
when fitting.
77
User Guide to MODDE
Autoscale modifier
In the PLS Scaling box, click Autoscale Modifier to change the modifier. Leaving the
modifier at the default=1 gives the same result as when selecting Unit Variance. Enter
a different value of the modifier and the response will be scaled to unit variance
multiplied by the value of the modifier.
Note: To keep a response out of the analysis set its autoscale modifier to zero
(i.e., enter 0 in the edit field).
Derived responses
A Derived response is a computed response as function of the factors and/or fitted
regular responses. When you add a derived response, you enter its formula. Derived
responses can be edited and deleted.
Derived responses are displayed in the responses spreadsheet. The values of the
derived responses are entered automatically in the worksheet when the model is fitted.
The derived responses are also available for all plots and list under the Worksheet and
Prediction menus and can be used as regular responses under these menus.
Note: Derived response values are only available after fitting (clicking Fit on
the Analysis menu) the model. When responses are included in the formula,
MODDE uses the fitted (predicted by the model) values of the responses in the
computation.
78
Responses
When clicking the Edit button, the Derived Response Wizard opens. The first page
contains information about derived responses. Select the Don’t show this page again
check box if you do not want to see this information again.
79
User Guide to MODDE
Clicking Next opens the next page in which you Enter the formula for the derived
response.
When you click Finish, MODDE parses the formula for correctness, and only
computes and displays the derived response in the worksheet, when you fit the model.
The derived response is added to the responses spreadsheet and the worksheet.
Note: Derived responses are deleted when regular factors are deleted, or
changed and when responses that are part of the derived response are deleted.
80
Responses
Note: To denote a set of variables you have to use the square brackets [ ], and
not regular parenthesis ( )
Functions
The functions available are Log10, Ln (natural log), and Exp (exponential).
81
User Guide to MODDE
Additional operators
The following additional operators apply to variables or sets of variables:
Avg(v[int1:int2]) Average of variables vint1... to vint2
Note: The parser is not case sensitive (t and T mean the same thing).
Linked responses
A Linked response is a response available in one investigation but fitted in another.
Linked responses are no longer available in MODDE.
Investigations containing linked responses are converted in MODDE 9 to instead hold
different models and/or worksheets. MODDE supports as many models in one
investigation as there are responses.
82
Responses
83
User Guide to MODDE
Adding a response
To add responses, double-click the last entry in the Responses spreadsheet window.
Enter the responses name, abbreviation etc, in the Response Definition dialog.
Modifying a response
The fields Name and Abbr. can be edited directly in the Responses spreadsheet.
To modify other fields for regular or derived responses, mark the desired response in
the responses spreadsheet and:
• Double-click it or
• Press ENTER.
The Response Definition dialog opens with the attributes of the response to modify.
Deleting a response
To delete one or more responses, mark the response(s) in the responses spreadsheet
and press the DELETE key on the keyboard or click Edit | Delete.
A dialog is displayed to confirm the deletion.
84
Constraints and inclusions
Constraints
A common problem is that experimentation may not be possible in some region of the
experimental space. For example it may not be possible, in an experiment, to have high
temperature and simultaneously low pH, and you want to cut-off the corner High temp,
Low pH. In MODDE this is solved by adding a Constraint using Design | Constraint.
A linear constraint is a function of the factors that specify a part of the experimental
region to be included or excluded.
The resulting experimental region is an irregular polyhedron. The corners of this region
are called the extreme vertices; they constitute part of the candidate set, i.e. a
discrete set of potentially good runs.
D-optimal designs are the only designs available when the experimental region is
constrained to an irregular polyhedron.
Constraints can be defined for quantitative or formulation factors.
Specifying constraints
Enter your constraints in the Constraints spreadsheet. The Constraints window is
opened by clicking Constraints on the Design menu.
In the upper part, the spreadsheet, you define each constraint (one per row) as a
mathematical relation. In the lower part, the graphical view, you can define constraints,
to be added to the upper part, geometrically. Such constraints may include two factors
only and are shown in the upper part after clicking the Add button.
85
User Guide to MODDE
Constraints supported
MODDE supports linear constraints, specified as exclusions, for quantitative process
factors or mixture factors Xk of the form
The first constraint specifies to exclude experimental runs where the sum of X3, X4,
and X5 is < 0.6
The second constraint specifies to exclude experimental runs where the sum of X4 and
X5 < 0.3
86
Constraints and inclusions
87
User Guide to MODDE
Note: Click a row in the spreadsheet defining a constraint in two factors and
MODDE displays the graphical constraint.
88
Constraints and inclusions
89
User Guide to MODDE
Inclusions
In MODDE designs can be augmented either using Complement Design (see the
section Complement Design in chapter File menu) or Inclusions.
Inclusions are extra runs that will be part of the worksheet. You can include a set of
experimental runs (inclusions), either at the end of the worksheet or to be part of a D-
Optimal design.
The Inclusions spreadsheet can be opened by clicking Design ⏐ Inclusions or by
clicking the import or Edit buttons in the Change D-Optimal settings page in the
Design Wizard.
Note: The inclusions are added to the worksheet only when you click the Add
to Worksheet button
90
Constraints and inclusions
Note: When generating D-Optimal designs, and the Include in design check
box is selected, the inclusions are a part of the design and included in the
number of runs.
91
User Guide to MODDE
Generating inclusions
Click Design | Inclusions to open the Inclusions spreadsheet, or click the Edit button
in the Inclusions area in the D-Optimal page in the design wizard.
In the inclusions spreadsheet you can add experiments by typing or pasting, by
importing from file, and by importing the current worksheet. You can do several
imports.
Importing worksheet
Click the Import Worksheet button to import the current worksheet to use as
inclusions.
Modifying inclusions
If the inclusions already exist, clicking Design | Inclusions or the Edit button in the D-
Optimal page in the design wizard, opens the spreadsheet with the inclusions for
editing.
To delete rows, mark them, press the DELETE key or right-click the spreadsheet and
click Delete.
92
Objective, model and design
Introduction
The Objective is the purpose for creating the design. MODDE recognizes two
objectives: Screening (first stage of an investigation when little is known) and
Response Surface Modeling (RSM) (optimization with the important factors.). The
Split Objective supports both screening and RSM. Paste Data defaults to the
screening objective with a linear model.
After defining your factors and responses, clicking Design ⏐ Objective opens the
design wizard which guides you through the selection of objective, design, and model
of the investigation.
The following sections describe the definition of the objective and selection of model
and design. See the Design Appendix chapter for details concerning the available
designs.
There are two dialogs associated with the menu item Objective:
• First: the Select Objective dialog to select the purpose.
• Second: the Select Model and Design dialog to select the type of design and
model.
93
User Guide to MODDE
Screening objective
Select the Screening objective when:
• You are starting an investigation and know little about the effects of the
factors on the response, the behavior of the response in the experimental
region, or the true size of that region.
• The goal is to reduce the number of factors to those with the largest effect on
the response.
This objective is available for all types of factors and factor combinations.
RSM objective
Select the RSM objective when:
• A lot is known about the investigation i.e. important factors, the size of the
region etc.
• The goal is to approximate the response by a mathematical model for the
purpose of prediction, optimization or finding a region of operability.
This objective is not available when all factors are qualitative.
Split Objective
Select the Split Objective when the investigation holds both process and mixture
factors AND you want to specify separate models for each.
If you want to specify one model for both mixture and process factors, select
Screening or RSM as objective.
The split objective is only available when there are both process and mixture factors
available.
94
Objective, model and design
Paste Data
Select Paste Data when you have the design and want to paste it instead of MODDE
creating one for you.
After selecting Paste Data and clicking Finish, the MODDE worksheet will expand
dynamically to fit the size of the pasted data. Click Analysis | Evaluate to view the
condition number of the current worksheet and model.
See also the Import design from file section in the File chapter.
Designs in MODDE
The design is the protocol for varying the factors in each experiment. Thus the design
is a set of experimental runs spanning the experimental region.
See the Design appendix for more details concerning the designs available in
MODDE.
95
User Guide to MODDE
Recommended designs
According to the selected objective and number of factors, MODDE recommends the
two most likely designs. Continue with the recommended design or select another one
by pointing and clicking or pressing the UP and DOWN arrow keys on the keyboard.
MODDE recommends, whenever possible, classical designs.
The recommendations are marked First and Second in the Recommendation column.
Note: By clearing the Show extended list of designs check box you can
display all supported designs in MODDE available for the defined factors and
selected objective.
Runs in design
In the Runs column the number of runs in the design is displayed. When there is a '+'
and/or a '-' sign after the number that means that the number of runs can be changed for
that particular design.
Model
MODDE supports polynomial models, such as linear, interaction, and quadratic. Third
order terms such as cubic or three factor interactions may be added to the model in
Design ⏐ Edit Model after the design generation.
The model for each design is listed in the Model column.
Screening models
Linear and Interaction models are appropriate for the screening objective. When the
model you select is:
• Linear MODDE generates the linear model. You may edit the model and
enter selected interactions.
• Interaction MODDE generates the full interaction model, i. e. all the two
factor interactions are included.
RSM models
Quadratic models are used for the RSM objective. For classical mixture designs cubic
designs are also available. When the model you select is:
• Quadratic MODDE generates the full quadratic model holding all two-factor
interactions and all the square terms of all the factors.
• Special cubic or cubic MODDE generates models accordingly. Such models
are only supported with mixture factors. Such models include all two-factor
interactions, all square terms, and some or all cubic terms.
Split models
When selecting the Split Objective, the model for the process factors and the mixture
factors can be specified independently of each other by clicking the Settings button.
96
Objective, model and design
Design runs
The Design runs box displays the number of runs for the selected design.
When there exists 2 or more fractional factorial designs of the same resolution, with
different number of runs, the number of runs, in the Runs column, is marked with a
“+”. MODDE defaults to the design with smallest number of runs. Use the Design
runs arrow to select the larger design.
For example, with the screening objective for 7 or 8 factors there exist two-resolution
IV design, one with 16 runs and the other with 32. MODDE selects the one with 16
runs. To select the design with 32 runs, click the Design runs arrow.
With D-Optimal designs the number of runs, in the Runs column, is marked with a ‘+’
and ‘-’ indicating that there exists smaller and larger designs.
When augmenting a design D-Optimally, the number of runs includes the number of
inclusions.
97
User Guide to MODDE
Center points
The Center points box displays the number of center points. MODDE always
recommends 3 center points. To change the number of center points:
• Type the desired number
• Click the Center points arrows and click a number.
• Mark the Center points box and press the up or down arrow keys on the
keyboard.
Replicates
The Replicates box displays the number of times to replicate the whole design
including center points. The default is '0', meaning that the design is not replicated.
Enter '1' here to replicate the design once.
To change the number of replicates:
• Type the desired number
• Click the Replicates arrows and click a number.
• Mark the Replicates box and press the up or down arrow keys on the
keyboard.
Total runs
After Total runs, the total number of runs included in the worksheet is listed and
includes: runs in the design plus center points and replicates.
98
Objective, model and design
Settings
For some designs the Settings-button is active.
Click the Settings button to:
• Edit the generators and/or model for fractional factorial designs of
resolution III, V, and V.
• Change the star distance for a CCC design.
• Specify the model when you have selected the split objective.
99
User Guide to MODDE
Note: The model and generators can be edited outside the design wizard by
clicking Edit | Model and Edit | Generators.
For more details see, the sections Model / reference mixture and Generators in the
Edit chapter.
100
Objective, model and design
Select the desired models and select the Add interaction between linear process and
mixture factors check box to add interactions between the process and mixture
factors.
Click the Edit model button in the D-Optimal page to edit the models further.
Description
Click the Description button, positioned below the Settings button, to display a short
description of the selected design. To hide the text, click the button again.
101
User Guide to MODDE
Blocks
MODDE supports Orthogonal Blocking for the 2 levels Factorial, Fractional factorial,
Plackett Burman, CCC, Box Behnken, and D-Optimal designs.
The maximum number of blocks supported by MODDE is 8, with a minimum block
size of 4.
Select the number of blocks to include in your design from the Blocks box.
Orthogonal blocking
The method of dividing experiments into blocks, so that the block effect is
uncorrelated with the main factor effects is called orthogonal blocking.
Orthogonal blocking is a way to deal with extraneous sources of variability that are not
included in the model. For example if one is making 32 experiments and the batches of
raw material are sufficient for 8 experiments, one would like to run the experiments in
blocks of 8 such as the variation between batches of raw material does not affect the
estimate of the main factor effects.
See also the Orthogonal blocking section in the Statistical appendix.
Block interaction
An interaction between a main effect and a block effect is called a block interaction.
When the design supports the interactions between the block effects and the main
effects, the Block interactions check box, in the Select the model and design page is
active. You can select the check box if you want to add the block interactions to your
model.
For details see the Orthogonal blocking section in the Statistical Appendix.
102
D-Optimal
103
User Guide to MODDE
This Change D-Optimal settings page consists of three sections relating to:
1. Design Generation Criteria
2. Design Alternatives
3. Candidate set
Design runs
Design runs is the number of runs the D-Optimal algorithm will generate, not
including the center points. You can change this number as desired. The smallest
number of runs accepted is the number of terms currently included in the model.
Model terms
The number of terms currently in the model is listed after Model terms. This number
is updated after changes in Edit Model.
104
D-Optimal
Note: When the investigation contains only mixture factors, the Edit Model
button is unavailable. The D-Optimal design is always generated from the full
model specified in the design page.
With investigations containing both mixture and process factors, you can only edit the
process factor terms and the interactions between mixture and process factors.
Potential terms
By default MODDE includes a set of potential terms, i.e. additional terms not included
in your model that might be important. The objective is to select a D-Optimal design
rich enough to guard for the potential terms. If you want your design to be just optimal
for your specified model, clear the Use potential terms box.
Inclusions
To use runs available from file as inclusions, click the Import button.
To edit the available inclusions, or paste/type runs to use as inclusions, click the Edit
button.
If you have specified runs as inclusions in the Design | Inclusion window prior to
entering the design wizard, the Include in design check box found under Inclusions, is
by default selected and the inclusions will automatically be part of the D-Optimal
design. Clear this check box if you do not want the inclusions to be part of the D-
Optimal design (but rather manually added at the end of the worksheet).
To add the inclusions after generating the worksheet, open the Inclusions under the
Design menu, and then click Add to worksheet.
Degrees of freedom
Number of Degrees of freedom of the residuals is calculated as:
Number of design runs – Model terms +1 (when you have center points)
The number of degrees of freedom recommended for D-Optimal designs in MODDE is
at least five.
105
User Guide to MODDE
Repetitions
In the Repetitions box, select the number of designs you want to generate with the
same number of runs, N. This will give a set of designs for each value of N.
If you want MODDE to only select balanced designs, select the Use balanced only
check box.
To be able to get a balanced design, the selected number of design runs must be a
multiple of the number of levels of the qualitative factor. The number of design runs
may be updated, if necessary, to be a multiple of the number of levels of the qualitative
factor.
106
D-Optimal
The candidate set can also be opened for editing by clicking Design | D-Optimal |
Candidate set.
107
User Guide to MODDE
D-Optimal results
When you click Next on the Change D-Optimal settings page, MODDE generates the
D-Optimal designs and displays them in the D-Optimal results page.
By default, the best design according to G-Efficiency is selected. Use the Auto-select
design by box to instead select the best design according to Determinant or
Condition number. Or select another design manually by marking the design.
Click any column header to sort the list.
To see the D-Optimal results as a plot, select the Display as plot check box.
The columns with the grid pattern represent the currently selected design.
108
D-Optimal
Right-click the plot and click Properties to open the D-Optimal Design Plot dialog
allowing:
• Selection of which of the criteria to display, G-efficiency, Determinant, or
Condition number under Show.
• Sorting on Type instead of Design (default) under Sort by.
• Selection of which layers to display when D-Optimal onion designs have
been generated in the Show Layer box.
Click Finish to generate the worksheet. Any already existing design and
worksheet will be deleted.
Click Design | D-Optimal | Generate to regenerate the D-Optimal designs as many
times as needed.
Note: The menu Design | D-Optimal | Generate is not available for onion
designs. To generate a new onion design, select Design | Objective.
Generate
Click Design | D-Optimal | Generate to open the Change D-Optimal Settings page
to re-enter the D-Optimal pages and generate a new set of D-Optimal designs.
See the section D-Optimal pages in the design wizard previously for more.
Candidate set
Click Design | D-Optimal | Candidate Set to open the Candidate set for viewing or
editing.
See the section Candidate set previously for more.
109
User Guide to MODDE
Evaluate
Click Design | D-Optimal | Evaluate to open the D-Optimal Results page to view the
other generated designs and their properties. Here you can also select one of the other
designs as your worksheet by marking and clicking Finish.
See the section D-Optimal Results previously for more.
Design plot
Select Design | D-Optimal | Design Plot to display the same information as Evaluate
in a column plot. MODDE displays G-efficiency, Log of the determinant of X'X, and
the condition number of X, for the D-Optimal designs.
When you generate several D-Optimal designs with different N (number of runs), you
can plot any of the D-Optimal criteria (G-efficiency, Condition No, Log of the
determinant of X'X) as a function of N. This plot can also be created from the Design |
D-Optimal | Evaluate and selecting the Display as plot box.
See the Design appendix for more information.
Note: All of the statistics available in the D-Optimal Design Plot are
computed from the runs selected D-optimally and do not include the possible
center points added to the worksheet. With mixture factors, the condition
number refers to the slack variable model with all mixture components scaled
orthogonal.
110
D-Optimal
Candidate set
The D-Optimal onion design in MODDE is created from a candidate set. The candidate
set can be created by MODDE, imported from one of the supported file formats, or
imported from SIMCA-P.
111
User Guide to MODDE
112
D-Optimal
When you click Finish, MODDE generates several D-Optimal designs in each layer
varying the number of runs by plus and minus 2, and displays the Onion D-Optimal
results.
The table on this page displays, for every layer, the generated designs statistics. By
default, in this table, the designs with the highest G-efficiency are selected.
You can select a different design in a given layer by marking it in the list or using the
Auto-select design by box and selecting a different criterion.
113
User Guide to MODDE
Candidate set
Click Design | D-Optimal | Candidate set to display the candidate set.
The columns in the spreadsheet are as follows when the candidate set was imported:
1. Design Run number. Corresponds with the experiment number in the
worksheet.
2. The Exp Name (experiment name) when available.
3. The Layer number (the innermost layer = 1).
4. The distance to the center of the multivariate space in Percent.
5. The design variables.
Runs in the candidate set that are not used to generate the D-Optimal design (not
included in the selected percentile ranges), are colored light gray.
When the candidate set is generated by MODDE the additional columns Layer and
Percent are unavailable. The layer belonging is displayed in the onion plots and when
listed (right-click and click Create List) the list is organized according to layer.
Evaluate
Click Design | D-Optimal | Evaluate to display the Onion D-Optimal results page.
You can use the table to select different designs, in selected layers.
See the D-Optimal Results section previously for more.
114
D-Optimal
Onion plots
There are two onion plots to visualize the candidate set and the selected D-Optimal
design, the Onion Plot and the Onion 3D scatter.
For the onion scatter plots, the property page has two tabs: Select Factors and Plot
Labels. Use Select Factors to select which factors to display on the X, Y, and Z-axes.
Use Plot Labels to select which labels to display in the plot.
Onion plot
To create the 2D onion plot, click D-Optimal on the Design menu, and then click
Onion Plot.
The onion plot is a 2D scatter plot of the candidate set. The candidate set runs are
colored by layer, and the selected design runs are black.
115
User Guide to MODDE
Onion 3D Scatter
To create the 3D Onion plot click D-Optimal on the Design menu, and then click
Onion 3D Scatter.
The Onion 3D scatter plot displays the candidate set colored by layer with the
selected design runs in black.
You can customize the plot using the plot settings. For more, see the Plots and List
chapter.
116
Design Wizard
Introduction
The Design Wizard opens when you click File | New, to create a new
investigation, and guides you from the start of the investigation to the generation of the
worksheet. The design wizard can also be opened by clicking Design | Design Wizard.
Exit the wizard at any time by clicking Finish.
The accelerator for the Design Wizard is CTRL+W.
Following is a description of the pages in the Design wizard.
Defining factors
The first page of the design wizard is the Define Factors page. On this page you can
define (enter new factors), modify, and/or delete factors.
117
User Guide to MODDE
All functionality available in the Factor spreadsheet described in the Factor chapter
is available in this page of the design wizard. Specific for the design wizard factor
page:
• The New and Edit-button open the Factor Definition dialog.
• The Delete-button deletes the factor(s) currently marked.
• Selecting the Place constraints on the experimental region box and
clicking Next opens the Constraints page of the design wizard. The
Constraints spreadsheet, including graphical constraints, is described in the
chapter Constraints and inclusions.
Note: The graphical constraint using the interface to define the constraint is
not available from the design wizard. If you want to use the graphical
constraint, click Design | Constraints
Clicking Next opens the Define responses page.
118
Design Wizard
Defining responses
In the Define responses page you can define (enter new response), modify, or delete
responses.
119
User Guide to MODDE
Clicking Next opens the Select the model and design page.
120
Design Wizard
Note: You can sort the Design list on any column by clicking on its header.
121
User Guide to MODDE
The selected design is highlighted and marked with the worksheet icon.
Note: Sort the list according to any column by clicking its header.
122
Worksheet
Introduction
After you have selected the model and design, MODDE generates the worksheet. The
worksheet is a spreadsheet used for storing the data (factors and response values) and
is where you enter the experimental results. The worksheet is by default sorted in
standard order.
Whenever you change the design (by editing the generators or selecting another
design) both the design and the worksheet are deleted and new ones are generated.
MODDE issues a message whenever the worksheet is regenerated.
The data in the worksheet are used when selecting to plot or list from the Worksheet
menu.
Worksheet menu
After the worksheet has been created and you have entered results, there are a number
of plots and lists available under the Worksheet menu.
All responses, regular and derived, are available in the plots and lists of the Worksheet
menu.
Worksheet button .
123
User Guide to MODDE
To re-randomize the run order after the worksheet has been created but before entering
results, on the Worksheet menu, click Set Run Order.
The run order can only be randomized for experiments with no results in the
worksheet.
Fully randomized
With RSM designs and Screening designs at more than 2 levels, Fully randomized is
the order in which you should perform the experiments and this is also the only
available order for these cases.
124
Worksheet
Strong curvature
If the Curvature Diagnostic Plot exhibits strong curvature as is shown above, you
should first re-measure the center point and re-do the plot. If the plot still exhibits
strong curvature, reduce the ranges of the factors by 2 / 3 and restart the project.
No curvature
If the Curvature Diagnostic Plot does not exhibit curvature, as the one below,
continue performing the rest of the experiments.
125
User Guide to MODDE
Click the Delete button in the dialog to remove factors or responses from the Y-
axis or from Series. To change what to display on the X-axis, mark and click the arrow
to add the new variable.
Click the Color by Variable tab to color by a factor or a response. With the scatter
plot created from the Worksheet menu you can also color by run order and experiment
number.
You can customize the 3D plot using the property page and plot settings. See the Plots
and List chapter.
126
Worksheet
Histogram
On the Worksheet menu click Histogram to display the histogram of the currently
selected response.
Select the desired response or add more responses by right-clicking the plot and
clicking Properties, or by making the selection in the Response box.
Transform in Histogram
You can transform a response by right-clicking the Histogram plot and clicking
Transform.
127
User Guide to MODDE
Descriptive statistics
Descriptive statistics can be displayed both as plot and in a list for all available
responses.
128
Worksheet
Correlation
The linear correlation coefficients R between all the terms in the model and all the
responses are displayed in the Correlation Matrix and Correlation Plot.
Process factors are transformed, scaled, and centered as specified in the
factor definition for MLR (default = orthogonal scaling). Responses are transformed as
specified in the response definition.
Formulation factors are always scaled orthogonally.
The value of the correlation coefficient R represents the extent of the linear association
between two terms. The value of R ranges from -1 to 1. When R is near zero there is no
linear relationship between the terms.
Correlation matrix
On the Worksheet menu, click Correlation, and then click Matrix.
Correlation coefficients above the threshold, between a term in the model and the
responses are colored green and those between terms of the model are colored red.
To change the threshold, colors, or number format, right-click and click Properties.
Correlation plot
On the Worksheet menu, click Correlation, and then click Plot.
The default plot displays the 10 largest correlation coefficients.
To change number of correlations to display or limit the number of correlations
according to a threshold:
1. Right-click the plot and click Properties.
2. Make the change, for instance click Show absolute correlations above
threshold and enter a value. With '0' all correlations are displayed
129
User Guide to MODDE
Replicate plot
In the replicate plot the values of the response are plotted vs. experimental runs
displaying the variation in the response for replicated experiments.
On the Worksheet menu click Replicate Plot to display the plot.
Select the desired response or add more responses by right-clicking and clicking
Properties, or by making the selection in the Response box.
Note: When the response has been transformed the Replicate plot displays the
back transformed values. To display the plot in the transformed metric, select
the Show transformed values check box in the Options tab in Properties.
Replicated experiments
MODDE checks the rows of all the factors (both included and excluded) in the
worksheet for replicates. Rows in the worksheet with the same factor values plus or
minus a tolerance are considered replicates.
The default Replicate tolerance is 0.1 (10%). You can change the Replicate
tolerance used in General Options on the View menu, tab General.
The replicates are used for the computation of the pure error and displayed on the same
Replicate Index in the Replicate Plot.
130
Worksheet
Worksheet spreadsheet
Description of the worksheet
Experiment number: Exp No start with number one and are assigned sequentially.
They cannot be edited.
Experiment name: MODDE assigns a default experiment name, Exp Name, on the
form Nxx, where xx is the experiment number. You may edit the name and enter your
own identification. The Plot Label named Experiment Name displays the text entered
under Exp Name when plotted.
Run order: The Run Order is the order in which the experiments should be
performed. MODDE suggests a randomized run order. Sort the worksheet according to
run order before performing the experiments.
Include or exclude: The fourth column labeled Incl/Excl indicates if the experiment is
included or excluded from the analysis. When the worksheet is generated all
experiments are marked Incl and are included in the analysis. To exclude an
experiment from the analysis select Excl in the worksheet.
Note: Excluded rows are excluded from the analysis for all responses. To
exclude the response value for only one response, right-click the cell and click
Exclude value(s).
Factors: In the Factor columns the factors are listed in original units.
Blocking: When you have selected Blocks in the Select the model and design page,
the column $BlockV displays which block each experiment has been assigned to.
For details on blocking, see the Orthogonal blocking section in the Statistical
appendix.
Responses: In the columns to the far right all responses are found. The response values
are listed in original units.
131
User Guide to MODDE
132
Analysis
Introduction
After entering the response values in the worksheet, analyze the resulting data in the
menu Analysis.
Use the Analysis Wizard to guide you through from raw data analysis to interpretation
and diagnostics.
You can also manually fit the model to the data by using either MLR (Multiple
Linear Regression) or PLS (Projection to Latent Structure) and clicking Fit. To review
the fit, display as lists or plots, the summary of the fit, the coefficients, the effects, and
the analysis of variance (ANOVA). Residual plots and the Box-Cox plot (only for
MLR) are available for diagnostic purposes.
133
User Guide to MODDE
Evaluate
Click Evaluate on the Analysis menu to display an evaluation of the current worksheet
and model. The table includes condition number, (number of) worksheet runs, (number
of) model terms, DF (degrees of freedom) residual, DF lack of fit, and DF pure error
individually for each response. These values differ between the responses when they
have different models or excluded values.
Condition number
The Condition Number is the ratio of the largest and the smallest singular values of X
(eigenvalues of X'X) where X is the extended design matrix. This condition number
represents a measure of the sphericity of the design (orthogonality). All factorial
designs, without center points, have a condition number of 1 and the design points are
situated on the surface of a sphere. For more see the Statistical appendix.
With factors orthogonally scaled all classical screening designs have a condition
number =1 without center points and <3 with center points. The condition number of
classical RSM designs varies according to the number of factors but remains <10.
134
Analysis
Analysis Wizard
The Analysis Wizard guides you through the main steps from review of the raw data
and the fit, to the diagnostics and refining of the model.
When opening the Analysis Wizard, the model is automatically fitted using the default
fit method.
In the Analysis Wizard each response is handled separately in a step wise manner
where you start looking at plots, pruning the model etc. for one response, and when
done you step over to the next response.
The following features are available as buttons in the Analysis Wizard:
• Display or hide Min, Target, and Max limits when available in the response
definition.
Note: All changes done in the Analysis Wizard take effect immediately.
Clicking Close closes the Analysis Wizard but all changes remain.
response, or click the button in the Standard toolbar. Clicking the Analysis
Wizard button opens the Analysis Wizard for the first response.
135
User Guide to MODDE
136
Analysis
Note: If your X matrix has a condition number > 3000, MODDE will only fit
the model with PLS and the condition number selecting MLR is displayed as
infinite.
Scheffé MLR
Select Scheffé MLR to fit the mixture data with a Scheffé type model. When you select
this fit method, the model is restored to its default specification. When your
investigation contains both mixture and process factors Scheffé MLR is unavailable as
such models are only available in MODDE for experiments with mixture factors only.
137
User Guide to MODDE
Note: The default method of fit with the Cox reference mixture model is PLS.
When the model obeys mixture hierarchy you can if you want fit the model
with MLR. When fitting the model with PLS, the condition number refers to
the X matrix, with unit variance coding.
138
Analysis
Note: R2 and Q2 provide the best summary of the fit of the model. R2 is an
overestimated and Q2 an underestimated measure of the goodness of fit of the
model
Saturated models
R2 and R2 Adjusted are not available for saturated models, i.e. models with DF
(degrees of freedom) equal to zero (see statistical appendices). However, with PLS Q2
can still be computed for saturated models and is hence displayed in plots and lists.
Statistics
In the plot footer MODDE displays:
N = Number of experimental runs.
DF = N-p (the degrees of freedom of the residuals).
Cond. no. = Condition number of the extended design matrix coded as selected in the
factor definition box: MLR scaling, or for PLS, unit variance.
2
R
The first column in the summary plot is R2 and is the fraction of the variation of the
response explained by the model:
R 2 = SS REG /SS
SSREG = the sum of squares of Y corrected for the mean, explained by the
model.
SS = the total sum of squares of Y corrected for the mean.
R2 overestimates the goodness of fit.
The R2 value is always between 0 and 1. Values close to 1 for both R2 and Q2
indicate very good model with excellent predictive power.
For details see the Statistical appendix.
You may select to plot R2 Adjusted instead of R2 by clicking it in the
Properties page or set it to be default displayed by clicking Investigation
Options on the View menu, tab R2.
R2 Adjusted: The fraction of variation of the response explained by the model
adjusted for degrees of freedom.
139
User Guide to MODDE
Q2
The second column in the summary plot is Q2 and is the fraction of the variation of the
response predicted by the model according to cross validation and expressed in the
same units as R2.
Q 2 = 1 - PRESS/SS
PRESS = the prediction residual sum of squares.
SS = the total sum of squares of Y corrected for the mean.
Q2 underestimates the goodness of fit.
The Q2 is usually between 0 and 1. Q2 can be negative for very poor models. With PLS
negative Q2 are truncated to zero for computational purposes. Values close to 1 for
both R2 and Q2 indicate a very good model with excellent predictive power.
For details see the Statistical appendix.
Model validity
The third column in the summary plot is the Model Validity and is a measure of the
validity of the model.
When the model validity column is larger than 0.25, there is no lack of fit of the model.
This means that the model error is in the same range as the pure error.
When the model validity is less than 0.25 you have a significant lack of fit and the
model error is significantly larger than the pure error (reproducibility).
A model validity value of 1 represents a perfect model.
Validity = 1 + 0.57647*log(plof)
where plof = p for lack of fit.
Reproducibility
The forth column in the summary plot is the Reproducibility which is the variation of
the response under the same conditions (pure error), often at the center points,
compared to the total variation of the response.
Reproducibility = 1 - (MS(Pure error)/MS(total SS corrected)).
MS = Mean squares, or Variance.
A reproducibility value of 1 represents perfect reproducibility.
140
Analysis
For every fitted response the plot displays R2 and Q2. The definition for R2 and Q2 is
the same as in the Summary of Fit Plot.
You may select to plot R2 Adjusted instead of R2 by selecting it in the Properties
page or setting it as default in View | Investigation Options, tab R2.
The condition number is calculated for the extended design matrix with the factors
scaled to unit variance.
141
User Guide to MODDE
This plot displays the R2 and Q2 per PLS component for the selected response. You can
change the selected response using the Response box. You may select to display R2
adjusted instead of R2 in the Property page and in the Investigation Options menu.
Saturated models
R2 and R2 Adjusted are not available for saturated models, i.e. models with DF
(degrees of freedom) equal to zero (see statistical appendices). However, with PLS Q2
can still be computed for saturated models and is hence displayed in plots and lists.
142
Analysis
Investigating diagnostics
Use the diagnostic plots and lists to find outliers, needed transformations etc.
143
User Guide to MODDE
Residuals
Under Analysis | Residuals three residual plots and one list are available.
144
Analysis
Right-click the plot and click Properties to select which variable to display on the x-
axis.
Residual list
To display the Residuals List, on the Analysis menu, click Residuals, and then click
List.
The number under the response name, is the experiment number.
Observed: The value of the response as listed in the worksheet.
Predicted: The predicted value for that observation.
Observed - Predicted: The residual for that observation.
Confidence Interval: The 95% confidence interval on the predicted value. You can
change the confidence level in the Property page and in Investigation Options.
145
User Guide to MODDE
146
Analysis
The Box Cox plot displays the maximum likelihood as a function of the power of the
transformation by plotting values of lambda, λ, vs. the maximum likelihood.
If the response values vary more than a magnitude of ten in the experimental domain, a
transformation is often recommended.
The maximum point on the Box Cox plot gives the value of (lambda, λ) for the
response transformation Yλ that gives the best fit of the model. This is the maximum
likelihood estimator for λ. For more, see the Statistical appendix.
147
User Guide to MODDE
The observed vs. predicted plot displays the plot of the observed values vs. the fitted or
predicted values. Plots with the points close to straight line indicate good models.
A regression line can be fitted by clicking the regression button on the Plot toolbar.
148
Analysis
The lack of fit plot compares the Lack of Fit (LoF) component to the pure
error component and displays a graph with 3 bars.
• SD-LoF: Shows the variation of the response due to the lack of fit of the
model (i.e. the model error) adjusted for degrees of freedom and in the same
units as Y. This is the square root of MS (mean square) lack of fit.
• SD-pe (Pure error): Shows the variation due to the replicated experiments
(observations) adjusted for degrees of freedom and in the same units as Y.
This is the square root of MS (mean square) pure error.
• SD-pe*sqrt(F(crit)): Shows SD pure error (second bar) multiplied by the
square root of the critical F.
The critical F is the value of the F-distribution over which SD LoF is statistically
significant at the 95% confidence level.
Hence, when the third bar is smaller than the first, the lack of fit is significant at
the 5% level (see statistical appendices).
149
User Guide to MODDE
ANOVA plot
To display the ANOVA plot, on the Analysis menu, click ANOVA, and then click
Anova Plot.
In the ANOVA plot the regression component is compared with the residual
component and 3 bars are displayed.
• SD Regression: Shows the variation of the response explained by the model,
adjusted for degrees of freedom and in the same units as Y. This is the square
root of MS (mean square) regression.
• RSD: Shows the variation of the response not explained by the model,
adjusted for degrees of freedom and in the same units as Y. This is the
residuals standard deviation.
• RSD*sqrt(F(crit)): Shows RSD (second bar) multiplied by the square root
of the critical F.
The critical F is the value of the F-distribution over which SD regression is statistically
significant at the 95% confidence level.
Hence, when the third bar is smaller than the first, the model is significant at the
5% level. For more see the Statistical appendix.
150
Analysis
ANOVA table
To open the ANOVA table, on the Analysis menu, click ANOVA, and then click
Anova Table.
The analysis of variance (ANOVA) table is displayed for the selected response.
If there are replicated observations, the residual sum of squares is further partitioned
into PURE ERROR and LACK OF FIT. A goodness of fit test is performed by
comparing the MS (mean square) lack of fit to the MS (mean square) pure error. See
the Statistical appendix for more information.
Replicated observations
MODDE checks the rows of the factors in the worksheet for replicates. Rows in the
worksheet with the same factor values plus or minus a 10% tolerance interval are
considered replicates and used for the computation of the pure error. The replicate
tolerance can be changed in General Options.
Note: The red coloring of the p-values always refer to the 95% resp. 5%
levels.
151
User Guide to MODDE
Coefficient plot
To create the coefficient plot, on the Analysis menu click Coefficients, and then click
Plot.
• The coefficient plot displays the coefficients, when changing from 0 to high,
for the selected response with the confidence interval as error bars. By
default, the coefficients refer to the data scaled and centered.
• You can select which type of coefficient to display from Properties. Select
Scaled and centered (default), Normalized, Unscaled, or PLS orthogonal
(only available for PLS).
• When you have confounded terms in your investigation these terms are
marked with a bracket #. Hoover over the column in the plot and you will get
information about which term(s) it is confounded with. This information is
available also in the coefficient overview plot.
152
Analysis
153
User Guide to MODDE
Coefficient list
To create the coefficient list, on the Analysis menu click Coefficients, and then click
List.
P-values signaling non significant coefficients at the selected confidence interval are
colored in red.
For the selected response(s), this list includes:
Terms: Under the response name the name of the terms included in the model are
listed. To display the list for another response, or more than one response, use the
Response bar or Properties.
Coefficients: Value of the coefficient.
Standard error: Standard error of the coefficient
P Value: Probability to get the displayed value for the coefficient if its true value was
zero.
Confidence interval: The 95% confidence interval on the coefficient value. To select a
different level for the confidence interval open Properties or Investigation Options.
154
Analysis
Note: When an expanded list of the qualitative levels is desirable, use the
coefficients plots and lists to display or list the coefficients for every level of a
qualitative variable instead of the effects plots and lists. Note that the effects
for linear and interaction models are twice the corresponding coefficients.
For mixture factors this plot displays the adjusted Cox effects (unavailable for Scheffé
models). This effect represents the change in the response values when component k
varies over its range, all other mixture factors kept in the same proportion as in the
reference mixture.
For details on how the mixture effects are calculated, see Statistical appendix, section
Mixture data in MODDE.
155
User Guide to MODDE
This plot, proposed by Daniel's in 1954, is based on the fact that if all estimated effects
were noise, they would have a normal distribution and when plotted on a normal
cumulative plot, would fall on a straight line. Hence effects significantly different from
zero (noise) will fall outside the normal line.
Note that this plot assumes independent effects, and that all estimable effects are
plotted. Hence, it is only relevant for screening designs with saturated models DF = 0.
Also for this plot to be meaningful, one needs models with at least 10 effects. If these
conditions are not met MODDE will warn you that this plot may not be statistically
correct.
Note: The normal probability of effects plot is not available with mixture
factors
156
Analysis
157
User Guide to MODDE
Interaction plot
To open the interaction plot, on the Analysis menu, click Effects, and then click
Interaction Plot.
To switch to another interaction and/or switch the factor displayed on the x-axis:
1. Right-click the plot and click Properties.
2. In tab Interaction Effects select the term in the Interaction term box and/or
select the factor in the Factor on X-axis box.
Note: This plot is only available for process factor. No interaction plots are
available for mixture factors.
When you select a 2 factor interaction, the predicted values of the response, when one
factor varies from its low to its high level, are plotted for both levels of the other factor,
all remaining factors in the design being set on their average.
If you have mixture factors in the model, these are all set at the standard reference
mixture.
158
Analysis
159
User Guide to MODDE
Effects list
To open the effects list, on the Analysis menu, click Effects, and then click List.
The values of the effects (twice the coefficients) are listed with their 95% confidence
interval sorted (in absolute value) in descending order.
160
Analysis
The VIP plot displays the VIP values as a column plot sorted in descending order.
The VIP list displays the sorted VIP values and the PLS coefficients for all responses
in the investigation.
161
User Guide to MODDE
162
Analysis
The loading plots display the correlation between the X variables T(X) and the Y
variables U(Y).
163
User Guide to MODDE
164
Prediction
Introduction
To make prediction using the model means:
• Using contour plots for interpretation.
• Making predictions in the prediction plots and list.
• Optimizing for selection of “best conditions”.
• Investigating robustness to disturbances using the Design Space.
The prediction menu includes:
• Contour Plot Wizard
• Prediction Plot Wizard
• Response Prediction Plot
• Prediction List
• Scatter Plot
• Sweet Spot Plot
• Optimizer
• Design Space Validation - only briefly described here but more thoroughly
in the Design Space chapter and Appendix D: Design Space.
Note: The Prediction menu applies to all responses, regular and derived.
165
User Guide to MODDE
Selecting responses
You can select to display all responses for the 2D contour plots but with more than 9
the plots become very small. The contours are not overlaid, but displayed next to each
other.
Select which response(s) to display by marking in Available responses and clicking
the => and/or marking in Selected responses and clicking <=.
Using constraints
Select the Use constraints box if you want the available constraints to be displayed in
the plot.
166
Prediction
Plot options
To select the resolution of the plot, that is, the grid calculated to create the contour plot,
click Plot Options. Note that you can type a resolution value here or click one of the
predefined resolutions.
You can also select to scale the subplots equally, to lock the contour levels, to produce
the plot with/without color and with/without contour level labels.
For more, see the Customizing contour plots sections in the Plots and lists chapter.
2D contour
The 2D contour plot displays the predicted response values, spanned by two factors, in
a response surface contour plot. For mixture the plot is spanned by three factors.
On the first page of the contour wizard, under Plot type, click Contour.
Under Inner plot type, if you have both process and mixture factors in the model,
click Process or Mixture to indicate which factors you want to display on the inner
axes.
Select which responses to display and click Next.
167
User Guide to MODDE
To hide the constants, right-click the window, and clear View Constants.
Zooming subplot
To zoom in on a subplot, click Subplot in the Zoom plot menu from the Plot toolbar.
168
Prediction
4D contour
The 4D Contour plot displays the predicted response values, spanned by two factors,
in 9 response surface contour plots in a 3x3 grid spanned by another two factors. For
mixture the plots are spanned by three factors.
On the first page of the contour wizard, under Plot type, click 4D Contour.
Under Inner plot type, if you have both process and mixture factors in the model,
click Process or Mixture to indicate which factors you want to display on the inner
axes.
Select which response to display and click Next.
169
User Guide to MODDE
170
Prediction
By default all plots are equally scaled, that is, the color coding is the same for all plots.
171
User Guide to MODDE
Response Surface
The Response Surface plot displays the predicted response values, spanned by two
factors, in a response surface plot. This plot is only available for models with two or
more process factors and can only display one response.
On the first page of the contour wizard, under Plot type, click Surface.
Select the responses to display and click Next.
To hide the constants, right-click the plot and clear View Constants.
172
Prediction
Plot type
To select to create a prediction plot, under Plot type click Prediction or Overlay
prediction on the first page of the prediction plot wizard.
Factor type
Under Factors at the X-axes, click the type of factor to vary: Process or Mixture.
You cannot vary simultaneously both process and mixture factors.
Selecting responses
You can select to display up to 3 responses for the prediction plot and 9 for the overlay
prediction plot. Select which response(s) to display by marking in Available responses
and clicking the => and/or marking in Selected responses and clicking <=.
173
User Guide to MODDE
Range or reference
You can select to keep the relative amount of the other mixture factors in the same
proportion as their ranges, rather than the reference mixture (default) by clicking
Adjusted to range.
174
Prediction
Factor type
The response prediction plot can be displayed for process and mixture factors although
only one at a time.
Process factor
When the selected factor is a process factor, all other factors are kept constant on their
averages.
Mixture Factor
When the selected factor is a mixture factor, the relative amounts of all other mixture
factors are kept in the same proportion as in the standard reference mixture. If no
standard reference mixture is specified, the centroid of the constrained region is used
as default.
Response prediction plot example
175
User Guide to MODDE
Prediction list
In the prediction list you can type or paste factor settings to make predictions. The
lower and upper level 95% confidence intervals for these predictions are also
calculated.
To open the prediction list, on the Prediction menu, click Prediction List.
To insert rows, right-click the list and click Insert Rows or press the DOWN arrow
key when on the last row.
If the model has been fitted the predictions are calculated automatically after entering
settings for all factors. If you do not want the predictions to be automatically updated,
right-click the list and clear Auto Update. To make predictions with auto update
cleared, right-click the list and click Update Predictions or press the F5 key.
To change the confidence level of the confidence intervals, right-click the list and click
Properties.
Note: With PLS and an X matrix with large condition number, the standard
error of predictions is computed and displayed instead of the confidence
interval.
Scatter plot
In the prediction scatter plot you can view your factors and predicted responses as 2D
and 3D scatter plots.
Note: The points displayed in the scatter plots are the points in the current
prediction list. The scatter plot is updated with new points when you enter
them in the prediction list.
To create a prediction scatter plot, on the Prediction menu, click Scatter Plot.
For 2D plots, select the variable to be plotted on the X-axis by marking it and clicking
the arrow ‘=>’. Mark the other variable(s) and click the arrow to Series.
For 3D plots, select the variables to be plotted on the X, Y, and Series axes.
Click the Delete button in the dialog to remove factors or responses from the Y-
axis or from Series. To change what to display on the X-axis, use the arrow to add the
new variable.
Click the Color by Variable tab to color by a factor or a response.
You can customize the 3D plot using the property page and plot settings. See the Plot
settings for 3D scatter and Onion 3D plots section in the Plots and Lists chapter.
176
Prediction
Plot type
To select to create the 2D sweet spot plot click Sweet Spot and to select the 4D with
up to four factors (5 with mixture) click 4D Sweet Spot. Click Next.
Response Selection
In the Response Selection page, enter the settings for the relevant responses.
Note: If you have entered Min and/or Max values in the response definition,
this page is automatically filled with those values. To update with updated
values from the response definition, click the Get Limits button.
For each response you have to select:
• To Include or Exclude each response under Incl/Excl.
• Which type of values, Value or Percent, you are entering under Min and
Max.
• Type the values that are of interest under Min and Max. MODDE has
entered the smallest and the largest values found in the worksheet. Change
them according to your desired sweet spot. You can also select or clear the
Use Constraints check box and change the resolution in the Resolution box.
177
User Guide to MODDE
Plot axes
Click Next to open the Axes and Constants page, or for 4D sweet spot, the 4D Axes
page.
Mixture factors can only be selected as inner factors in a 4D plot.
Select two process factors for the axes under Factors at the plot axes, or three mixture
factors
For the 4D, select two process factors for the outer axes. The sweet spot contours are
plotted for the low, middle, and high levels of these factors.
178
Prediction
179
User Guide to MODDE
Optimizer
You can use the optimizer to find an experimental area limited both by factor settings
and response criteria.
To open the optimizer, on the Prediction menu click Optimizer. MODDE opens a
dialog box with 3 spreadsheets.
180
Prediction
Response spreadsheet
In the response spreadsheet all the responses used in your model are available. Before
starting the optimization you must select the Criteria for your response(s), weights and
limits.
If you have specified Min, Target, and/or Max in the response definition, these
specifications are copied to the Optimizer response section. If no Min, Target, or
Max settings are defined in the response definition, the response will be by default
Predicted.
Note: You can fetch updated limits from the response definition by right-
clicking and clicking Update limits from response definition.
181
User Guide to MODDE
Weight
In the column Weight you can enter a number between 0.1 and 1 reflecting the
importance of the response. Default is 1 indicating that all responses are of equal
importance.
The weight will change the optimization objective described in the first section of this
chapter as follows:
• Weight = 1 for all responses results in a limit optimization (default).
• Weight = 0.2 for all responses results in a strict target optimization.
• Individual weights for the responses results in an optimization where the
responses with higher weights are favored in the quest for all responses to
reach inside the limits.
For more information see Appendix C - Optimizer.
182
Prediction
Run list
The run list includes the factors, responses, iterations, Log(D), and DPMO. The D in
Log(D) is a normalized distance to the target. See Appendix C: Optimizer for more
information.
In the run list above, row 6 has a DPMO = 0 meaning that with disturbances + 5% on
the factor settings will give a solution inside the specifications. Row 1 has
approximately the same Log(D) but a DPMO = 5600 indicating that a small
disturbance in the factor settings from this point will result in some hits outside the
specifications. Therefore row 6 is preferable.
To use absolute limits, select the Absolute response limits check box above the run
list, or right-click the window and click Properties to open the Optimizer Properties,
in tab Options, select the Use absolute response limits check box.
183
User Guide to MODDE
Sensitivity analysis
The Analyze sensitivity check box is by default selected to display complementary
information about the proposed optimal factor settings. The sensitivity is expressed as
DPMO (Defects Per Million Opportunities) outside specifications, based on Monte
Carlo simulations and including prediction error. Note that DPMO is an estimate of the
result after the number of simulation performed; 10 000 simulations will give a fair
estimate.
Optimizer Properties
In the Options tab optimizer settings are specified.
Feature Description Default
Use absolute When selected, only the runs where all responses are predicted Not
response limits inside the specified limits are displayed in the run list. selected.
When selected, the DPMO is calculated for runs where all responses
are predicted inside the specified limits. The Sensitivity Range for
Calculate DPMO the factors specifies the range used in the DPMO calculations. Selected.
This is a sensitivity analysis that indicates if a solution is sensitive or
insensitive to small changes.
Include Model When selected, the sensitivity analysis (DPMO) includes the model
Selected.
Error error in addition to the sensitivity range.
Include weight in
When selected, the weight specified in the response spreadsheet is Not
calculation of
included when calculating Log(D). selected.
Log(D)
184
Prediction
Optimizer buttons
There are seven optimizer buttons: Generate start runs, run optimizer, generate start
points from selected, contour plot wizard, copy to predictions, sweet spot plot wizard,
and analyze design space.
The Design Space, opened by clicking the Analyze Design Space - button is described
in the Design Space chapter.
185
User Guide to MODDE
Start the optimizer by clicking the Run Optimizer (play) button or right-click
the list and click Run Optimizer.
The optimizer runs to completion and displays the results as:
• Final factor settings.
• Predicted responses values.
• Number of iterations.
• Log(D), i.e., the logarithm of the overall distance to Target. The smaller the
Log(D) the better is the result. At Log(D) = -10 all predicted responses are
on target.
• DPMO, estimated number of hits outside specification on one million
simulations. A smaller DPMO means that the solution is less sensitive to
disturbances.
Restarting the optimizer
After convergence, you can always click the Run Optimizer (play) button to restart
the optimizer. It restarts from the displayed resulting runs of the previous search. If you
do not want to continue with the resulting runs click the Generate new starting points
or Generate starting points around selected button to specify other start runs.
Mark the run, and click the Generate start points from selected button or right-
click the list and click New Runs from Selected.
In the dialog box displayed the run you marked is the default run to optimize around.
To switch to another run, enter another run in the Center around optimizer run box.
Enter the percent of the factor range in the Factor range box. The percentage entered
here is used to calculate the new high and low limits for the start runs.
186
Prediction
Mark the best run and click the Contour Plot Wizard button to open the
Contour Wizard Setup dialog box. You can also open the wizard by right-clicking
the list, clicking Plot and selecting Contour Plot Wizard.
187
User Guide to MODDE
settings, click the Copy to Predictions button or right-click and click Copy to
Predictions.
188
Prediction
To create the sweet spot plot click the Sweet Spot Plot Wizard button . You can
also open the sweet spot plot by right-clicking, clicking Plot and then clicking Sweet
Spot Plot. In the dialog that opens you can select to create the 2D sweet spot plot by
clicking Sweet Spot or 4D by clicking 4D Sweet Spot.
The selected run is displayed in the plot as lines from the axes with arrows pointing
toward the position of the selected run.
189
User Guide to MODDE
190
Design Space
Introduction
The calculation of the design space is a search function that expands the possible factor
ranges from a setpoint (optimum) to the largest possible range where all response
predictions are still within the specifications.
Predictions in the design space are done with Monte Carlo simulations. The resulting
distribution of predictions simulates a real situation with a random combination of
factor setting disturbances within a given range.
There are two Design Space features in MODDE:
• Predictive Design Space Estimation for optimization.
• Design Space Validation for robustness testing.
DS is used as an abbreviation for Design Space in this user guide.
191
User Guide to MODDE
192
Design Space
Factor spreadsheet
In the factor spreadsheet the following functionality is available for each factor:
Item Description Setting
The yellow lines are the factor settings of the selected optimal run in the optimizer.
The red lines are the Low and High factor settings of the experimental region.
Estimated The black T bar represents the region of acceptable variability valid for that factor
acceptable when all other factors are locked at the optimum. Valid means that all predictions of
range the responses are within the specifications. No model error is considered in this
search.
The blue region represents the 95%* part of random factor variability with normal
distribution where all predictions are within the specifications.
*95% is the default. In the Design Space Properties you can change to 99%.
193
User Guide to MODDE
Response spreadsheet
In the response spreadsheet the following functionality is available for each response:
Item Description Default
'Minimize’, ‘Maximize’,
‘Target’, and ‘Predicted’ as
specified in the optimizer.
Criterion What the algorithm is aiming for. When the search does not
reach inside the limits for a
response, the Criterion is set
to 'Not Met'
The yellow line represents Target value for the responses as specified in the
Predicted Optimizer.
response The red lines are the specification limits for each response as specified in Optimizer.
profile The faded green region represents the probability of a prediction for a random
distribution of factor settings in the given ranges (low-optimum-high), the DS.
* See table in the DPMO and Cpk section in Appendix A: Statistical notes.
factor.
• Properties opens the Design Space Properties dialog. For more see the
Design Space Properties sub section.
194
Design Space
Parameters section
In the Parameters section the settings used in the design space estimation are
displayed and can be adjusted.
Option Description Default
Acceptable range Number of simulations done in each step of the search for the
20 000
simulations acceptable range for the factors.
Response profile Number of simulations for the final predictions of the response
100 000
simulations profile.
Defines the Low and High values in such a way that 95%, in the
Limits at 95 or 99 %
default case, of the Monte Carlo simulations are inside the limits.
confidence for 95%
Selecting 99% widens the included area by lowering the Low limit
normal distribution
and increasing the High limit in the factor spreadsheet.
195
User Guide to MODDE
Show section
In the Show section what to display and how can be adjusted.
Option Description Default
Columns shown in the response part of the DS. The By default Responses, Min,
Response available fields are: Responses (response names), Min, Target, Max, Criterion,
fields Target, Max, Criterion, Cpk, DPMO, Average, St.Dev, DPMO, and Predicted
%Outside, Median, 1st Quartile, 3rd Quartile, and response profile are
Predicted response profile. displayed.
Factor histogram
When double-clicking the estimated acceptable range for the factors, the Factor
Histogram opens. This histogram displays the distribution of the factor settings used
for the simulation.
196
Design Space
Response histogram
The result of the Monte Carlo simulations can be presented as frequency histograms
by:
• Double-clicking Predicted response profile.
197
User Guide to MODDE
198
Show
Introduction
On the Show menu the following items are available: Objective list, Design Matrix,
Design Region plot, Confoundings list, D-Optimal Summary table, Model list, and
Reference Mixture list.
Objective
To open the Objective list, on the Show menu click Objective. The Objective list
displays the selections made in the Objective pages of the design wizard
199
User Guide to MODDE
Design matrix
To display the Design Matrix, on the Show menu, click Design Matrix.
The Design Matrix displays the experimental plan in coded unit for quantitative
factors, as in the worksheet for qualitative factors and in pseudo components for
formulation factors.
If you have qualitative factors at more than two levels, you can select to display the
design with the qualitative factors extended.
To display the design matrix with the qualitative factors extended:
1. Right-click the matrix and click Properties.
2. Click Extended – shows all settings and click OK.
3. Optionally switch back by clicking Regular – shows all orthogonal
settings.
By default the Design Matrix is derived from the worksheet, and reflects any changes
(excluded runs, changed values, additional runs etc.,) done to the factor part of the
worksheet after it’s generation.
To display the original design matrix generated by MODDE:
1. Right-click the matrix and click Properties.
2. Click The design as generated by MODDE and click OK.
3. Optionally switch back by clicking The current Worksheet scaled and
centered.
200
Show
Design region
Open the Design Region plot to get an overview of your experimental plan.
To open the Design Region plot, on the Show menu, click Design Region.
Use the Response box or right-click the plot and click Properties to switch to another
response.
Hint: When you want to view 3 factors in 3D, use the Scatter Plot found under
the Worksheet and Prediction menus.
201
User Guide to MODDE
Confoundings
Open the Confoundings list to see which terms that are mathematically identical in the
current design. For instance, in the example below the term M*TH is included in the
model, but the effect of this term is confounded with the effect of H*S. This means that
using this design there is no way of telling whether the coefficient displayed for M*TH
reflects M*TH, H*S, or a mixture of both.
To list the confoundings click Confoundings on the Show menu.
For factorial designs resolution III or IV, the Confoundings list displays the
confounding pattern for the complete interaction model.
In the Term column, the background of the terms included in the model is colored.
In the coefficient plots and lists, confounded terms are marked with a bracket #-sign.
202
Show
Summary D-Optimal
To displays a list summarizing the generation of a D-Optimal design click D-Optimal
Summary on the Show menu.
Under Candidate set the properties of the used candidate set are listed in the form of
Extreme Vertices, Edge points, Centroid of high dim. Surfaces, and Total Runs.
Under D-Optimal the properties of the selected design are listed in the form of
Objective, Model Type, Potential Terms, Number of Inclusions, Constraints,
Design Runs, Selected Design Number, G-Efficiency, log(Det. of X' X), Norm.
log(Det. of X' X), and Condition number.
203
User Guide to MODDE
Model
On the Show menu click Model to display the selected model.
If you have qualitative factors at more than two levels and want to display the model
with the qualitative factors extended:
1. Right-click the Model list.
2. Click Properties.
3. Click Extended – Show all settings.
Reference mixture
To display the reference mixture used in the analysis of the design click Reference
Mixture on the Show menu.
To change the reference mixture, click Model / Reference Mixture on the Edit menu
and click the Reference Mixture tab.
204
Plots and lists
Introduction
You can customize the plots in MODDE using the Plot Settings page. You can change
the properties of plots and lists from the property page accessed by right-clicking the
plot or list and clicking Properties.
Most plots can be transformed into lists by right-clicking the plot and selecting Create
List.
Properties page
Opening property page by right-clicking
To open the property page of a plot or list, right-click the plot or list and click
Properties.
205
User Guide to MODDE
Note: When displaying multiple plots, opening the Properties page displays
the properties for the first displayed response.
Plot settings
Use Plot Settings to customize the most common attributes of the axes, of the plot area
and of the header and footer.
There are two different types of plot settings dialogs in MODDE:
• Plot settings for 3D scatter plot and contour, response surface, and sweet spot
plot where all options are in the same dialog.
• Plot settings for other plots are divided in Axis, Header and Footer, and
Plot Area pages.
Open Plot Settings by double-clicking the plot or right-click and click Plot Settings,
then when applicable click Axis, Header and Footer, or Plot Area.
After customization, most features can be saved as default plot settings in two ways:
• Click the Save Settings button in the Plot Settings dialog.
• Right-click the plot and click Save as Default Plot Settings.
206
Plots and lists
Axis
To change the settings of the plot axes, right-click the plot, click Plot Settings, and
then click Axis. This dialog is also opened if you double-click one of the axes.
All of the properties apply to the selected axis: X, Y, or Y2 (the second Y-axis).
In this dialog you can change general settings, such as scale, gridlines, titles, and fonts
by clicking the appropriate tab.
Scale
To change the scale, enter the Minimum and Maximum data values and the value to
Increment by. You can specify the spacing of the minor tick marks and the number of
Decimal places to display on the tick label.
Click the Recalculate Scale button to restore the Minimum and Maximum values.
Miscellaneous
By default the axes are in normal scale. You can change them to Logarithmic scale or
Values in reverse order by selecting the corresponding check box. The axes are
default displayed. To hide an axis, clear the Show axis check box.
The Always recalculate scales check box is by default selected. If you do not want
MODDE to recalculate the scale you should clear the check box.
207
User Guide to MODDE
Gridlines
In the Gridlines tab, select the Show gridlines check box to display the grid lines of
the selected axis. The grid lines are placed on the tick marks. Select the Pattern,
Color, and Width of the gridline.
Title
In the Title tab, select to show or hide the axis title, change the text and its orientation.
208
Plots and lists
Font
In the Font tab, select the font for both the axis titles and the tick mark labels.
For how to customize the axes of the 3D scatter plots, see the section later in this
chapter.
For how to customize the header and footer of the 3D scatter plots, see the section later
in this chapter.
209
User Guide to MODDE
Plot Area
To change the settings of the interior of the plot, right-click the plot, click Plot
Settings, and then click Plot Area. This dialog is also opened if you double-click the
plot area.
In this dialog you can change symbol style, line style, data labels, and font by clicking
the appropriate tab.
Symbol style
Select the shape, color and size of all the symbols in a series.
Line style
Click the Line Style tab to change the properties of lines in line plots.
Data labels
Click the Data Labels tab to change the attributes of all the labels on data points inside
the plot area. Color, alignment, or orientation (Rotation) applies to all the labels in a
given series.
Font
Click the Font tab to select the font for all the data labels.
210
Plots and lists
Background
Click Background, found under Series to the left in the dialog, to change the borders
of the window, plot, and chart areas.
Legend
Click Legend, found under Background to the left in the dialog, and select or clear the
Show Legend check box to show or hide the legend. Here you can also change text
placement, color, background and border styles of the legend.
211
User Guide to MODDE
Column style
When you have a column plot you can change the pattern, color, column width, and
overlap for each or all series in the Format Plot Area dialog.
For how to customize the plot area of the 3D scatter plots, see the section later in this
chapter.
212
Plots and lists
Resolution
Resolution is the precision used when creating the contour plot. Selecting a higher
resolution requires more calculations and is therefore more time consuming.
To select the resolution of the plot, click one of the predefined resolutions or type a
resolution value here.
Use color
By clearing the Use color check box you can choose to display the contour and surface
plots without colors.
213
User Guide to MODDE
Axis
Click Axis to change settings of the axes. For more, see the Plot settings for 3D
scatter and Onion 3D plots section later in this chapter.
Header
Click Header or Footer to customize the positioning and coloring. For more, see the
Plot settings for 3D scatter and Onion 3D plots section later in this chapter.
Regions
Use the Region style page to customize how the area outside the factor setup is
visualized. By default, this area is shaded white.
Labels
Click Labels to:
• Clear the Show label check box to hide the plot labels.
• Select the position of the plot labels in the Anchor box.
• Select the Connect with attach point check box when the label is separated
from the line (using Offset) and you want a line between the label and the
point.
• Customize the Text and Background color and Transparency of the plot
label.
• Select the Is visible check box under Border to display a border around each
label. Enter custom values in the Margin and Width fields.
• Click the Font tab to customize the font.
Contour
In Contour Level Colors in the Plot Settings page you can customize levels and
colors according to the below.
The default plot displays a number of contour levels. To display fewer or more contour
levels, type a new value in the Number of Levels box. In Individual level colors the
new number of levels will be displayed.
Note: To for instance display the contour levels for 10, 20, 30 …100 you need
to type Min=0 and Max=110 in Color level range, and in the Number of
Levels box type 10.
214
Plots and lists
After Color range on the contour levels the color buttons, defining the color scale
used in the plot, are positioned.
To change from the default colors, click the respective buttons, and click the color to
use.
To change the color of one of the available levels under Individual level colors, mark
the level and click the Color button, and then click the new color.
To remove one of the available levels, mark it and click the Remove button.
You can also add levels by typing a value in the field under the Add button, and then
clicking the Add button.
Click the Contour Line Style tab to customize the contour line color, width, and
pattern.
Axis
Click Axis to change settings of the axes. For more, see the Plot settings for 3D
scatter and Onion 3D plots section later in this chapter.
Header
Click Header or Footer to customize the positioning and coloring. For more, see the
Plot settings for 3D scatter and Onion 3D plots section later in this chapter.
Contour
In Contour Level Colors in the Plot Settings page you can customize the colors
according to the below.
215
User Guide to MODDE
The default plot displays up to 5 colors in the legend. These levels are by default
colored from blue to green where green represents the sweet spot.
To change the color of one of the available levels:
1. Under Individual level colors, mark the level and click the color button.
2. Then click the new color.
To change the range used, click the respective color buttons and click the new color.
Cube
Click Cube to:
• Clear the Color cube sides check box to make the cube sides transparent.
When the Color cube sides check box is selected the sides are colored in the
color specified in Side color.
• Customize the Side color.
• Select/clear the Show gridlines and Gridlines on minor ticks check boxes
to show/hide the gridlines.
• Change the Line width, Pattern, and Line color to specify the properties of
the gridlines.
216
Plots and lists
Axis
Click Axis to change settings of the axes. For each axis you can select:
• To display the annotation of the axis.
• The minimum and maximum values of the axis and the step size.
• To autoscale the axis.
• The axis title.
In the Axis General tab you can:
• Customize the axes color and size.
• Customize the annotations color and distance from the axis.
• Select/clear the Show axis arrows check box to display arrows/regular axes.
• Select the axis arrow color.
• Select the level of Transparency for the axes and annotation respectively.
• Adjust autoscaling properties.
In the Ticks tab you can:
• Customize the tick's direction and size for normal and minor ticks.
• Clear the Show ticks and Show minor ticks check boxes to hide the ticks.
In the Font tab you can select the font and size of the annotation and titles of the axes.
217
User Guide to MODDE
Header
Click Header to:
• Select/clear the Is visible check box under Header to show/hide the header.
• Select the position of the header in the Anchor box.
• Customize the Border and text color of the header.
• Customize the Background color of the header.
• Select/clear the Is visible check box under Border to show/hide the border.
• Customize the Margin and Width of the border.
In the Font tab you can select the font and size of the header.
Legend
Click Legend to:
• Select/clear the Show legend check box to show/hide the legend.
• Select the position of the legend in the Anchor box.
• Select the orientation of the legend in the Orientation box.
• Select the text adjustment of the legend in the Text adjust box.
• Customize Border and Text color and Background color of the legend.
• Select/clear the Is visible check box under Border to show/hide the border.
• Customize the Margin and Width of the border.
Footer
Click Footer to:
• Select/clear the Is visible check box under Footer to show/hide the footer.
• Select the position of the footer in the Anchor box.
• Customize the Border and text color of the footer.
• Customize the Background color of the footer.
• Select/clear the Is visible check box under Border to show/hide the border.
• Customize the Margin and Width of the border.
In the Font tab you can select the font and size of the footer.
218
Plots and lists
Symbols
Click Symbols and the individual series to change the properties of the symbols. When
you mark the individual series you change the properties of that series only. When you
mark Symbols you change the properties of all series. In these pages you can
• Select the Shape of the symbol.
• Customize the Color of the symbol.
• Select the Size of the symbols.
• Select the Draw style of the symbol, where the choices are Fill, Wireframe,
and Point. Fill is the default.
• Select the Line style when the Draw style is Wireframe.
Labels
Click Labels to:
• Select/clear the Show label check box to show/hide the plot labels.
• Select the position of the plot labels in the Anchor box.
• Select the Connect with point check box when the labels are separated from
their position (using Offset) and you want a line between the label and the
point.
• Customize the Offset in X, Y, and Z space of the plot labels.
• Customize the Text color of the plot label.
Colors
Click Colors to:
• Change the Highlight Style displayed when hovering over a point. For more
about the content on this page, see the Symbols topic above.
• Change Marking Color for the marked points.
219
User Guide to MODDE
220
Report generator
Introduction
MODDE has an automatic report generator. You start the report by clicking File |
Generate HTML Report. A dialog opens where you can select the MODDE default
template, or any other report as a template.
In the report generator, basic formatting functionality is available for writing text.
Plots, lists, and model results of MODDE can be added to the report at any time. These
items are added to the report as placeholders.
A placeholder stands in the place of contents which MODDE will provide, let it be a
plot, list or any text or number.
The placeholders enable you to use the same report, as a template, in different
investigations. You can then edit the text, and just click the Update Report button to
update all MODDE results from the current investigation.
If you have saved a report with the same name as the investigation, MODDE default
suggests opening that report.
When your investigation has not been fitted MODDE will ask to fit the investigation so
it can automatically fill all the placeholders. You can prevent this question if you select
under View | General Options, to automatically fit investigations when opened.
221
User Guide to MODDE
File menu
On the File menu you find the general Windows commands New, Open, Save As,
Print Preview, Print and Exit. Additionally you find Continue Edit Report With,
Templates, and View in Browser.
222
Report generator
Templates
You can save templates, restore templates and add or remove custom templates by
clicking Templates on the File menu.
Click:
• Save as Default Template when you have changed/created a template
according to your wishes and want to use it as the default template next time
you generate a report.
• Restore Default Templates if you have made changes to the default
templates and want to remove those changes.
• Save as Custom Template when you have changed/created a template
according to your wishes and want to save it to be able to use it again.
Custom templates will be listed in the “Select template or open existing”
dialog and when you click Insert | Template.
• Add / Remove Custom Templates when you want to add an already created
template or remove one of your custom templates.
View in browser
To view the current report in your default internet browser, on the File menu click
View in Browser. This works with Internet Explorer and Mozilla Firefox.
223
User Guide to MODDE
Edit menu
On the Edit menu you find the standard commands Undo (CTRL+Z), Redo, Copy
(CTRL+C), Paste (CTRL+V), Paste Unformatted (CTRL+SHIFT+V), Clear (press
the Delete key), Select All (CTRL+A), and Find (CTRL+F).
View menu
On the View menu you can select to hide or show the following:
• Generate Report toolbar opened/closed by clicking View | Toolbar |
Generate Report. For more, see the Generate report toolbar section later
in this chapter.
• Format toolbar opened/closed by clicking View | Toolbar | Format. For
more, see the Format toolbar section later in this chapter.
• Status bar which displays an explanation when you point to a button. To
show/hide it, click View | Status Bar.
• Placeholders window which displays the placeholders that can be added to
the report. For more see the Placeholders window section.
• Properties window. The properties of plots and images can be customized in
the Properties window. To display the window click Properties. For more
see the Properties window section.
• Click Customize to customize toolbars and menus as in Office 2003 and
later. Find a short introduction in the Customize section in the View chapter.
Insert menu
Use the Insert menu to insert a Hyperlink, Image, File, or Template in the current
report. Here you also can toggle the Grab Plot or List mode explained below.
Hyperlink
To insert a hyperlink, select text in the report and then click View | Hyperlink, also
available by right-clicking the report and clicking Insert Hyperlink. In the dialog that
opens, enter the address of your hyperlink the URL field.
Image
Click View | Image to insert a picture in the report. Also available by right-clicking the
report and clicking Insert Image.
Click the Browse button to find your file.
File
Select View | File to insert a Web page file (*.htm, *.html), a Text file (*.txt), or a
picture file (*.jpg, *.png, *.gif, *.bmp.)
Template
Clicking View | Template opens a dialog where you can select a custom template to
insert in the report. Templates must first have been added using File | Templates |
Save as Custom Template.
224
Report generator
Format menu
On the Format menu you can customize Font and Styles and Formatting.
Tools menu
On the Tools menu, find the commands Update Report, Update Placeholder, Show
All Placeholders, Show Placeholder, Remove All Placeholders, and Remove
Placeholder.
Update Report
Click Update Report to update all placeholders with the plots and lists of the current
investigation.
Update Placeholder
Click Update Placeholder to update the marked placeholder.
Show Placeholder
Click Show Placeholder to show the underlying placeholder of marked plot, list, or
item.
Remove Placeholder
Click Remove Placeholder when you do not want a certain item, plot, or list to be
updated. Remove Placeholder is also available as a button on the Generate Report
toolbar.
225
User Guide to MODDE
Help menu
To access the FAQ of the report generator, click Welcome Page and FAQ on the
Help menu.
New
Click the arrow next to the New button and the following commands are displayed.
• Click New on the menu displays the generate report dialog from which you
can select which report/template to use or to open an existing
report/template.
• Click New Blank Report to start a new report with no text.
• Click New From Default Template if you have created a template that you
have saved as default template. Save a report as a default template by
clicking Templates on the File menu and then clicking Save as Default
Template. Umetrics’ default template is used if no other template has been
specified.
For more on the Placeholder buttons see the Tools menu section.
For more on the Grab Plot or List button see the Insert menu section.
For more on the Continue Edit With button see the File menu section.
Format toolbar
The Format toolbar is the standard toolbar for formatting text with three additional
226
Report generator
Placeholder window
Open the Placeholder dockable window by clicking View | Placeholders.
227
User Guide to MODDE
Properties window
Open the Properties dockable window by clicking View | Properties.
In the Properties window you can change the default plot size and to save a plot as
.png, .bmp, or .jpg.
Properties for placeholders are displayed when you click the placeholder. You can
change the properties of the placeholder in the Data field in the properties window.
Click the Data field to view a description on how you can change the properties of the
current placeholder.
228
Help
Introduction
MODDE’s help is based on this user guide. The user guide documents are transferred
in to a compiled HTML file. To read the Help file Internet Explorer version 4.0 or
higher must be installed but does not need to be your default browser.
HTML help
The HTML help file is installed to include interactive help and stand alone where the
program is installed.
Open the help by:
• Clicking Contents and Index from the Help menu.
Manage Licenses
If your company has a license server that handles you licenses, follow the instructions
delivered with the delivery letter.
If you choose to activate later, click Manage Licenses on the Help menu and then
follow the instructions delivered with the delivery letter.
229
User Guide to MODDE
About MODDE
To find the version number of MODDE, on the Help menu, click About MODDE.
230
Appendix A: Statistical notes
Fit methods
MODDE supports Multiple Linear Regression (MLR) and PLS (Projection to Latent
Structures) for fitting the model to the data.
231
User Guide to MODDE
Note: When the models for the responses are different, PLS fits each response
separately.
PLS has been extensively described in the literature and only a brief description is
given here.
PLS finds the relationship between a matrix Y (response variables) and a matrix X
(predictor or factor variables) expressed as:
Y = XB+E
The matrix Y refers to the characteristics of interest (responses). The matrix X refers to
the predictor variables and to their square or/and cross terms if these have been added
to the model.
PLS creates new variables (ta) called X-scores as weighted combinations of the original
X-variables: ta = Xwa, where wa is the combinations weights. These X-scores are few,
often just two or three, and orthogonal (independent). The X-scores are then used to
model the responses.
With several responses, the Y-variables are similarly combined to a few Y-scores (ua)
using weights ca, ua = Yca. The PLS estimations are done in such a way that it
maximizes the correlation, in each model dimension, between ta and ua, One PLS
component (number a) consists of one vector of X-scores (ta), and one of Y-scores (ua),
together with the X and Y-weights (wa and ca).
Hence the PLS model consists of a simultaneous projection of both the X and Y spaces
on a low dimensional hyper plane with new coordinates T (summarizing X) and U
(summarizing Y), and then relating U to T. This analysis has the following two
objectives:
• To well approximate the X and Y spaces by the hyper-planes
• To maximize the correlation between X and Y (u and t).
Mathematically the PLS model can be expressed as:
X = TP' + E
Y = TC' + E
232
Appendix A: Statistical notes
Geometrically, we can see the matrices X and Y as n points in two spaces, (see figure),
the X-space with p axes, and the Y-space with m axes, p and m being the number of
columns in X (terms in the model) and in Y (responses).
X3 Y3
t1
u2
t2
u1
X2 Y
2
X1 Y1
t1
u
1
233
User Guide to MODDE
Model
You may edit the model and add or delete terms for individual responses. You may add
up to third order terms (cubic terms, or 3 factors interaction).
If your design is singular with respect to your model, MODDE will fit the model with
PLS, and MLR will not be available.
Hierarchy
MODDE enforces hierarchy of the model terms. You cannot delete the constant term.
You can only delete a linear term if no higher order term containing the factor is still in
the model.
234
Appendix A: Statistical notes
Scaling
Scaling X
When fitting the model with multiple linear regression, the design matrix X is scaled
and centered as specified in the factor definition box: MLR scaling. If the choice is not
orthogonal, the condition number will differ from the one displayed when clicking
Analysis | Evaluate.
When fitting the model with PLS the X matrix is always scaled to unit variance.
If warranted, the scaled X matrix is extended with squares and / or cross terms
according to the selected model.
The choices of scaling are:
(x denotes the original factor value and z the scaled one)
Orthogonal scaling:
zi = (xi - M)/R
Where M = Mid-range, R = Range/2.
Mid-range scaling:
zi = (xi - M)
Unit variance scaling:
zi = (xi - M)/s
Where m = average, s = standard deviation computed from the worksheet.
Note that Orthogonal and Mid-range scaling are only available with MLR.
MODDE default scaling for MLR is the orthogonal scaling.
Scaling Y
The matrix of responses Y, when fitting the model with PLS, is by default scaled to
unit variance. You can modify the unit variance scaling weight by using the PLS
scaling box in the factor definition. With MLR the Y's are not scaled.
235
User Guide to MODDE
Condition number
The condition number of the orthogonally scaled and centered extended design matrix
using the SVD (Singular value decomposition) is computed when clicking Evaluate on
the Analysis menu. The X matrix is taken from the worksheet. The calculation depends
on fit method (MLR, PLS) and which factors are involved.
236
Appendix A: Statistical notes
Note: With formulation factors and fitting with PLS, the condition number is
computed by excluding the factor with the largest range, and scaling the
remaining ones orthogonal. When fitting with MLR the condition number is
computed without centering and scaling the factors.
Missing data
Missing data in X
Missing data in X are not allowed, and will disable the fit. This does not apply to
uncontrolled X-variables. For MLR experiments with missing in an uncontrolled factor
results in excluding that row in the calculations while PLS can handle this.
N-value
The N-value used in ANOVA, and for the computation of R2 adjusted, is the actual
number of non-missing observations (experiments) for each response-column. This N-
value and DF = N-p are displayed at the bottom of the ANOVA plots and lists, and on
all residual plots, including observed vs. predicted Y.
237
User Guide to MODDE
ANOVA
The analysis of variance (ANOVA) partitions the total variation of a selected response
SS (Sum of Squares corrected for the mean) into a part due to the regression model and
a part due to the residuals.
SS = SSregr + SSresid
If there are replicated observations (experiments), the residual sum of squares is further
partitioned into pure error SSpe and Lack of fit SSlof.
SSresid = SSpe + SSlof
DFresid = (n - p)
SSpe = ∑ ( eki - ek )2
DFpe = ∑ ( nk - 1 )2
DFlof = n - p - ∑ ( nk - 1 )2
1. where the ∑ loops over ki resp k.
2. n = number of experimental runs (excluding missing values)
3. nk = number of replicates in the kth set
4. p = number of terms in the model, including the constant
5. ek = average of the nk residuals in the kth set of replicates
6. j = jth residual in the kth set of replicates
A goodness of fit test is performed by comparing the MS (mean square) of lack of fit to
the MS of pure error:
Two ANOVA plots are displayed:
1. The regression goodness of fit test
2. The LoF goodness of fit test
238
Appendix A: Statistical notes
Q2
Q2 = (SS - PRESS)/SS
with
(Yi − Y$i )2
PRESS = ∑ 2
i (1 − hi )
where hi is the ith diagonal element of the hat matrix:
X(X'X)-1X'
R2
R2 = (SS - SSresid)/SS
R2 adj = (MS - MSresid)/MS
MS = SS / (n - 1)
MSresid = SSresid / (n - p)
RSD = Residual Standard Deviation = √MSEresid
Degrees of freedom
MODDE always computes the real degrees of freedom RDF of the residuals:
RDF = n - p - ∑( ni - 1 )
where the ∑ loops over i
n = number of experimental runs
ni = number of replicates in the ith set
p = number of terms in the model, including the constant
Saturated models
When RDF = 0 the model is saturated, and MODDE does not compute or display R2,
R2 Adjusted or Q2 when fitting the model with MLR. With PLS, only Q2 is computed
and displayed.
Singular models
Singular models (condition number > 3000) are only fitted with PLS.
If p > n - ∑( n i - 1 ) , the degrees of freedom of the residuals are computed as:
239
User Guide to MODDE
Coefficients
Scaled and centered coefficients
The regression coefficients computed and displayed by MODDE refer to the centered
and scaled data. You may also select to display the “unscaled and uncentered”
coefficients.
Normalized coefficients
In the overview plot, to make the coefficients comparable between responses, the
“centered and scaled” coefficients are normalized with respect to the variation of Y.
That is, they are divided by the standard deviation of their respective Y's.
Confidence intervals
Confidence intervals of coefficients and predictions are computed using the total
number of observations, regardless of missing values for PLS models with identical
models for all responses. For MLR and PLS with different model for the responses, the
total number of observations is the number of non-missing values per response. This
total number of observations is displayed as N at the bottom of all other plots and lists.
The approximation for PLS with identical models is possible because the confidence
intervals computed with the regression formulas are somewhat too large because the
PLS solution is a shrunk estimator with smaller prediction errors than those of
regression. Hence a small number of missing elements in Y does not make the PLS
confidence intervals larger than those computed with the regression formulas and the
total number of observations.
240
Appendix A: Statistical notes
a -1 -1 -1
b 1 0 0
c 0 1 0
d 0 0 1
The coefficients of these expanded terms are given as the coefficients for level 2 (b), 3
(c), and 4 (d) of C, while the coefficient for level 1 (a) is computed as the negative sum
of the three others. MODDE displays all the four coefficients in the coefficient table
but notes that they are associated with only three degrees of freedom.
Residuals
Raw residuals
The raw residual is the difference between the observed and the fitted (predicted) value
ei = Y i – Ŷ i
The raw residuals are displayed in the residual lists.
Standardized residuals
The standardized residual is the raw residual divided by the residual standard deviation
ei / s (s = RSD)
These are MODDE default for PLS Residual plots.
241
User Guide to MODDE
Predictions
For X matrices with condition number < 3000, both MLR and PLS computes a
confidence interval for the average predicted y:
Yi + √hi * RSD * t(α / 2, DFresid)
hi is the ith diagonal element of the hat matrix: X(X'X)-1X'
For X matrices with condition number > 3000 and for all Cox mixture models, PLS
computes only the standard error of the average predicted Y:
SE(Y) = √[(1 / N) + t'0 * (T'T)-1* t'0] * RSD
PLS plots
Both scores and loading plots are available:
Plot loadings
WC plots (PLS)
Plots of the X- and Y-weights (w and c) of one PLS dimension against another, say,
no.'s 1 and 2, show how the X-variables influence the Y-variables, and the
correlation structure between X's and Y's. In particular one better understands how the
responses vary their relation to each other and which ones provide similar information.
Plot scores
TT, UU, and TU plots (PLS)
The tt and uu plots, of the X- and Y-scores of, say, dimensions 1 and 2 (i.e. t1 vs. t2, and
u1 vs. u2), can be interpreted as windows into the X- and Y-spaces, respectively,
showing how the design points (experimental conditions, X) and responses profile (Y)
are situated with respect to each other. These plots show the possible presence of
outliers, groups, and other patterns in the data.
The tu plots (t1 vs. u1, t2 vs. u2, etc.) show the relation between X and Y, and display the
degree of fit (good fit corresponds to small scatter around the straight line), indications
of curvature, and outliers.
PLS coefficients
PLS computes regression coefficients (Bm) for each response Ym expressed as a
function of the X's according to the assumed model (i.e. linear, linear plus interactions
or quadratic,). These coefficients are (columns of B) computed as:
B = W(P'W)-1 C'
W and C are (p*A) and (m*A) matrices whose columns are the vectors wa and ca.
p = number of terms in the model
m = number of responses
A = Number of PLS components
242
Appendix A: Statistical notes
243
User Guide to MODDE
244
Appendix A: Statistical notes
k = rk*tk
rk = Uk - Lk
tk = bk/(T-sk)
where:
rk is the range of factor k
tk is the total Cox effect
T is the mixture total. In most cases T=1.
bk is the unscaled uncentered coefficient
sk is the value of the factor at the reference mixture
The Effect Plot is only available for screening designs using the Cox model.
Main effect plot
For a selected mixture factor Xk, this plot displays the predicted change in the response
when Xk varies from its low to its high level, adjusted for all other mixture factors, that
is, by default, the relative amounts of all other mixture factors are kept in the same
proportion as in the standard reference mixture (MODDE does not check if the other
mixture factors are kept within their ranges).
For example, if the Main effect of the mixture factor X1 is being displayed, when X1
takes the value x1, the other mixture factors are assigned the values: xj = (T - x1)* (sj /
T - s1).
Sk are the coordinates of the standard reference mixture. The standard reference
mixture is the one used in the model.
You can change this default and select to have all other mixture factors kept in the
same proportion as their ranges (this ensures no extrapolation).
Interaction Plot
Interaction plot is not available when you only have mixture factors.
245
User Guide to MODDE
Linear models
Σbksk = 0 (1)
Quadratic models
Σbksk = 0 (1)
Σckjbkjsk = 0 for k = 1,,,,q (1) and for j = 1,,,,q (2)
Here ckj = 1 when j ≠ k and ckj = 2 when k = j.
and sk are the coordinates of the standard reference mixture.
Note: In MODDE, with linear and quadratic models obeying the mixture
hierarchy, (i.e. terms constrained by (1) or (2) can only be removed as a
group, and not individually), by default the PLS coefficients are always
expressed relative to a stated standard reference mixture.
With models containing terms of the third order (cubic), or disobeying mixture
hierarchy, no constraints are imposed on the PLS coefficients. The coefficients are in
this case, the regular PLS coefficients and the reference mixture is implicit and results
from the projection.
246
Appendix A: Statistical notes
Quadratic
Scheffé‚ bk = Cox (PLS) b0 + bk + bkk
Scheffé‚ bkj = Cox (PLS) bkj - bkk - bjj
Scheffé models
The Scheffé models are only fitted with MLR and only the main effect plot is
available.
Scheffé models are only available for investigation with all mixture factors.
ANOVA
As described by Snee in “Test Statistics for Mixture Models” (Technometrics, Nov.
1974), the degrees of freedom in the ANOVA table are computed in the same way as
with the slack variable model.
247
User Guide to MODDE
Note: When the model contains terms of order 3, or contains qualitative and
formulation factors, the PLS coefficients are not adjusted relative to a stated
standard mixture.
MODDE plots
All MODDE plots are available when you have both mixture and process factors. For
both the Main Effect and prediction plots, when you select to vary a process factor, all
of the mixture factors are set to the values of the standard reference mixture. When you
select to vary a mixture factor, process factors are set on their average and the other
mixture factors are kept in the same proportion as in the standard reference mixture or
their ranges.
Optimizer
The Optimizer uses a Nelder Mead simplex method with the fitted response functions,
to optimize an overall desirability function combining the individual desirability of
each response.
For details on how to use the Optimizer, see the Appendix C: Optimizer chapter.
Desirability
For every response y, the desirability function is computed as follows:
f(g(y)) = 100*(e^(λ*g(y)) - 1)
where g(y) = 100*((y -P)/(T - P))
T, L and P are defined as follows:
T = User desired Target
L = User defined worst acceptable response value(s)
P = Worst response value computed from the starting simplex. P is never closer to the
Target than the L(s).
When the response is to be maximized, L is the smallest acceptable value, when the
response is to be minimized L is the largest acceptable value. When the response is to
be on Target, the user gives the smallest and largest acceptable values.
When the response is to be minimized, we must have T < L and when the response is
to be maximized we must have T > L.
For responses to be on target the user must supply lower and upper limits such as
L(lower) < T < L(upper)
L is generated internally when not supplied by the user.
λ is a scaling parameter computed as follows:
⎡ 100 ⎤
ln ⎢
(100 − Limit ) ⎥⎦
λ=− ⎣ ( L − P)
100
(T − P)
248
Appendix A: Statistical notes
Overall desirability
The overall desirability f(ds) is a weighted average of the individual desirability
function. The weights, denoted w, are the user entered weights between 0.1 and 1,
reflecting the relative importance of the responses.
⎡ ⎛ yi − T ⎞ ⎤
2
⎢ ∑ wi ⎜ ⎟ ⎥
⎢ ⎝ T − L⎠ ⎥
D = log 10
⎢ M ⎥
⎢ ⎥
⎣ ⎦
M = number of responses.
D = -10 when all responses have reached T.
D is not used in the optimization but is displayed as Log(D) in the run list.
Starting simplexes
The optimizer starts 8 simplexes from 8 starting runs selected as follows:
• The first four are from the corners of a 23-1 design in the factors with the
largest coefficients for the first response.
• The fifth is the overall center point.
• The last three are the 3 “best” runs from the worksheet considering only the
first response and the predicted values.
Duplicated start points will not be shown.
The user can modify these runs or add own.
Each simplex is generated from the starting run by adding an additional run for each
factor with an offset of 20% of the distance from the center to the maximum value, the
other factors being kept at the same values. A check is made that all runs are within the
defined experimental region.
249
User Guide to MODDE
Sensitivity Analysis
Sensitivity Analysis is an estimate of how sensitive the proposed factor settings are to
small disturbances. The sensitivity analysis is done with Monte Carlo simulations
(MC) and displayed as DPMO (Defects Per Million Opportunities outside
specifications). Normally distributed random disturbances of + 5% of the range of each
factor are applied and result in an estimate of the total number of DPMO for all active
responses. The user can modify the sensitivity range and the number of MC’s.
DPMO=0 means that the proposed factor settings represent a robust point of operations
for factor disturbances of + 5%.
Orthogonal blocking
When you cannot perform all of the experiments in a homogeneous way, randomizing
the run order of the experiments may not be sufficient to deal with the extraneous
sources of high variability. You may want to run the experiments in homogeneous
groups, i.e., blocks, in such a way that the external source of variability does not
influence the effects of the factors.
For example, suppose you are running a full factorial with 5 factors and 32 runs, and
the batch size of raw material allow you to perform only 8 runs per batch. You may
want to run your experiment in 4 blocks, each composed of 8 runs using homogeneous
material.
The method of dividing 32 runs in 4 blocks of 8 runs, each such as the difference
between the blocks (the raw material) does not affect the estimate of the factors, is
called Orthogonal Blocking.
MODDE supports orthogonal blocking for the 2 level factorial, fractional factorials,
Plackett Burman, CCC, and Box Behnken designs.
MODDE can also block D-Optimal designs. These designs are more flexible with
respect to the number of blocks and the block size, but the blocks in D-Optimal design
are not usually orthogonal to the main factors. The only restriction with D-Optimal
designs is that the number of runs must be a multiple of the block size.
Note: Blocking introduces extra factors in the design, hence reduces the
degree of freedom of the residuals, and the resolution of the design. You
should only block when the extraneous source of variability is high and
cannot be dealt with by randomizing the run order.
Block interaction
An interaction between a main effect and a block effect is called a block interaction.
When the design supports the interactions between the block effects and the main
effects, the Block interactions check box, in the Select model and design dialog in
the design wizard is active. You can select the check box if you want to add the block
interactions to your model.
250
Appendix A: Statistical notes
– – Block 1
+ – Block 2
– + Block 3
+ + Block 4
When you select Show | Design Matrix, the design matrix is displayed in coded units
including the blocking factors.
When the worksheet is generated, the blocking factors are recoded and the model is
reparameterized. Rather than keeping the d blocking factors, such as 2d = k (the
number of blocks), MODDE generates one qualitative variable called $BlockV, with k
levels called B1, B2 ...Bk.
251
User Guide to MODDE
252
Appendix A: Statistical notes
It is not always possible to generate a balanced D-Optimal design with respect to the
blocking factor. In this case you may want to change the model, the number of blocks,
or generate an unbalanced design.
253
User Guide to MODDE
Normal
Triangular
254
Appendix A: Statistical notes
The low and high factor settings are the distribution boundaries. For a normal
distribution 95% of the distribution is found within the boundaries, by default. The
automatic search procedure expands the distribution for each factor until one or more
response limits are exceeded according to the specified DPMO target. The automated
search procedure is symmetrical around the setpoint but random for the factors.
where
USL = Upper Specification Limit.
LSL = Lower Specification Limit.
μ = predicted average.
σ = estimated standard deviation for predictions.
255
User Guide to MODDE
1 1350 0.13
1.3 48 0.0048
1.4 13 0.0013
2 0.0010 9.86588E-08
For more information about Cpk (Process Capability Index) search in Wikipedia.
256
Appendix B: Designs
257
User Guide to MODDE
The default generators used by MODDE for fractional factorial designs are those
recommended by Box, Hunter and Hunter (page 410). You may edit and change the
generators in the Generator dialog available on the Edit menu. When you update the
confounding, MODDE will warn you if some of the effects in your model are
confounded with each other, i.e. if your model is singular.
D-Optimal designs
D-Optimal designs are computer generated designs that maximize the determinant of
the X'X matrix, X being the extended design matrix.
D-Optimal designs are available for all objectives.
For more see the D-Optimal designs section later in this chapter and see also the D-
Optimal chapter.
Onion designs
Like regular D-Optimal designs, D-Optimal Onion designs can be used both in
screening and in RSM with quadratic models. The Onion designs comprise layers of
designs, usually D-Optimal, where the outermost layer determines which type of model
(screening or RSM) that the Onion design supports. For more see the D-Optimal
onion designs section later.
258
Appendix B: Designs
Rechtschaffner designs
Rechtschaffner designs are orthogonal, saturated fractions of resolution V of the 2n and
3n factorial designs. They allow the estimation of all main effects and all first order
interactions without confounding. They are saturated designs, with no degrees of
freedom remaining for the estimation of residuals and diagnostics.
The 2n Rechtschaffner designs are well suited when the objective is screening, with 6
or more factors, and little knowledge about the importance of each individual first
order interaction. In this case it is of interest to estimate all first order interactions,
unconfounded, and then eliminate the insignificant (small) ones, hence recovering
some degrees of freedom for diagnostics and residual analysis.
The required number of runs N for the 2n Rechtschaffner designs with k factors is:
N =1 + k + k(k - 1)/2
It is recommended to add 3 to 4 center points to these designs.
RED-MUP designs
The RED-MUP designs are custom designs developed for the use with 96 well plates
(see figure) and larger (384, 1536, etc.). These are widely used platforms for
experimentation in biochemistry, microbiology, pharmaceutical development, etc.,
with some special properties (buildable and extendable to other factor intervals). The
RED-MUP designs consist of two sub-designs corresponding to the vertical and
horizontal directions of the plates, i.e., 8 and 12, respectively, for 96-well plates. The
total design is made by multiplying the two sub-designs together. Hence, this total
design supports a model with all interactions between the factors in the sub-designs,
plus from each sub-design, the main effects, and when these sub-design so support,
interactions, and quadratic effects.
Below, we use n1 and n2 for the number of rows and columns in the plate, i.e., 8 and
12 in a 96 hole plate. A 96-well plate can handle from 5 full “RSM factors” up to 18
factors for a stretched screening situation.
The layout of a 96 well plate has 8 rows and 12 columns. Hence, the vertical direction
has n1=8, and the horizontal direction has n2=12.
If both sub-designs support only main effects, using for example Plackett Burman sub-
designs, up to n1+n2-2 factors can be investigated, i.e., up to 18 factors in a 96 well
plate. Such a sparse design without center points is not recommended. More reliable
designs with center points in the larger sub-design would allow n1-3 + n2-5 = n1+n2-8
factors, i.e., 12 factors for a 96 well plate.
259
User Guide to MODDE
When specifying the RED-MUP design, it is important to distribute the factors over the
two sub-designs (vertical and horizontal) so that (a) the actual experimental protocol
remains simple and doable, (b) the sub-designs and final design make
chemical/biological/engineering sense, and (c) a-priori interesting interactions and
higher order terms can be estimated. Note that all the interactions between each of the
factors in the vertical design and each of the factors in the horizontal designs can
always be estimated. Hence, factor pairs for which interactions are expected should be
split into the two sub-designs. Then their interaction can always be estimated,
regardless of choice of sub-design.
Special designs
When selecting to create a RED-MUP design, there are special designs for the 96 well
plates (8 x 12) which aim to make better use of the plate.
RSM designs
RSM designs are used in later stages of an investigation to develop more elaborate
models (quadratic) in the few important factors, usually not more than 5 or 6.
MODDE supports the following RSM designs:
D-Optimal designs
D-Optimal designs are computer generated designs that maximize the determinant of
the X'X matrix, X being the extended design matrix.
D-Optimal designs are available for all objectives.
For more see the D-Optimal designs section later in this chapter and see also the D-
Optimal chapter.
260
Appendix B: Designs
Onion designs
Like regular D-Optimal designs, D-Optimal Onion designs can be used both in
screening and in RSM with quadratic models. The Onion designs comprise layers of
designs, usually D-Optimal, where the outermost layer determines which type of model
(screening or RSM) that the Onion design supports. For more see the D-Optimal
onion designs section later.
Rechtschaffner designs
Rechtschaffner designs are orthogonal, saturated fractions of resolution V of the 2n and
3n factorial designs. They allow the estimation of all main effects and all first order
interactions without confounding. They are saturated designs, with no degrees of
freedom remaining for the estimation of residuals and diagnostics.
The 3n Rechtschaffner designs are well suited for the RSM objective with 6 or more
factors as they require fewer runs than the classical CCC or CCF non saturated designs.
The intent with these designs is to estimate quadratic terms but performing fewer runs
than with CCC or CCF. Eliminating insignificant terms, after performing the
experiments, results in recovering some degrees of freedom.
The required number of runs N for the 3n Rechtschaffner designs with k factors is:
N = 1 + 2k + k(k - 1)/2
It is recommended to add 3 to 4 center points to these designs.
RED-MUP designs
The RED-MUP designs are custom designs developed for the use with 96 well plates
(see figure below) and larger (384, 1536, etc.). These are widely used platforms for
experimentation in biochemistry, microbiology, pharmaceutical development, etc.,
with some special properties (buildable and extendable to other factor intervals). The
RED-MUP designs consist of two sub-designs corresponding to the vertical and
horizontal directions of the plates, i.e., 8 and 12, respectively, for 96-well plates. The
total design is made by multiplying the two sub-designs together. Hence, this total
design supports a model with all interactions between the factors in the sub-designs,
plus from each sub-design, the main effects, and when these sub-design so support,
interactions, and quadratic effects.
The RED-MUP designs are well suited for the RSM objective with up to 5 or 6 factors.
The intent with these designs is to get a precise model that can be used for optimization
and for detailed understanding.
The maximum number of “RSM factors” depends on the sizes of the sub-designs. An 8
run sub-design, e.g., a Doehlert design with 2 center points, supports 2 “RSM factors”
(1 constant, 2 linear, two quadratic, and one interaction terms), and a 12 run sub-
design, e.g., a three level Rechtschaffner design with 2 center points, supports 3 “RSM
factors” (1 constant, 3 linear, 3 quadratic, and 3 interaction terms) for a total of 5
“RSM factors” for a 96-well plate.
Mixed objective
Since the RED-MUP designs are constructed from two sub-designs, one of these can
be an RSM design and the other a screening design. In such a case the objective is said
to be mixed.
261
User Guide to MODDE
Special designs
When selecting to create a RED-MUP design, there are special designs for the 96 well
plates (8 x 12) which aim to fill up the plate.
Doehlert designs
The Doehlert designs are quadratic RSM designs with some special properties
(buildable and extendable to other factor intervals). They allow the estimation of all
main effects, all first order interactions, and all quadratic effects without confounding.
They are saturated designs with similar properties to the CCF and CCC designs.
Geometrically they are polyhedrons based on hyper-triangles (simplexes), with a
hexagon in the simplest two-factor case.
Doehlert design in 2 factors with 6 runs + center points, can be extended to a new
design by adding 3 experiments. Usually also one or two new center points are added
in the new design (i.e., in the figure the right-most point in the old design).
The Doehlert designs are well suited for the RSM objective with up to 5 or 6 factors
(respectively 33 and 45 runs with 3 center points). The intent with these RSM designs
is to get a precise model that can be used for optimization and for detailed
understanding.
The required number of runs N, except for replicated center points, for the quadratic
Doehlert designs with k factors is:
N = 1 + k + k2
It is recommended to add 3 to 4 center points to these designs.
262
Appendix B: Designs
Formulation factor
Formulation factors are the usual mixture factors used in formulations with specifically
defined experimental ranges. Most mixture experiments have only formulation factors.
Filler factor
The presence of filler is typical of certain types of simple mixture experiments. For
example in a synthesis the solvent is typical filler, as is water in a juice punch. A filler
is a mixture component, usually of little interest, making up a large percentage of the
mixture, and added at the end of a formulation to bring the mixture total to the desired
amount.
It is recommended to define a mixture factor as filler when all three conditions below
are fulfilled:
• The factor is always present in the mixture.
• The factor accounts for a large percentage of the mixture and there is no
restriction on its range. It is added at the end to bring up the mixture total to
the desired amount (usually 1 when no mixture factors are kept constant).
• You are not interested in the effect of the filler per se.
When you specify a filler factor, MODDE checks that these conditions are met and
defaults to a slack variable model, with the filler factor omitted from the model.
Use
All mixture factors are controlled or constant. The Uncontrolled option is unavailable
for both formulation and filler factors.
Formulation factors can be defined as Constant when you want to keep them constant
in the experiment.
263
User Guide to MODDE
When mixture factors are constant, the mixture total T = 1 - Sum (constant mixture
factors). When no formulation factors are defined as constant, the mixture total has to
be equal to 1. MODDE issues an error message and stops whenever the mixture total is
not equal to T or 1.
Scaling
Mixture factors are always unscaled when you fit the model with MLR. When you fit
the model with PLS, all mixture factors are scaled to unit variance.
Note: When the mixture region is regular, mixture factors are first
transformed to pseudo components, and then scaled with PLS models.
Mixture constraint
In a mixture experiment the mixture total (i.e. the sum of all the mixture factors in the
experiment) is equal to a constant T. The mixture Total T is generally equal to 1 when
no mixture factor is kept constant. This mixture constraint implies that the mixture
factors are not independent, and this collinearity has implications on the mixture
experimental region, the mixture designs, and the mixture model formulation.
264
Appendix B: Designs
Note: You can select to have the analysis done on the mixture factors
expressed in original unit under the Analysis menu.
Screening designs
MODDE provides three variants of the axial design. Axial designs locate all the
experimental points on the axis of the simplex and are recommended for screening, see
Snee (references).
Standard Axial (AXN)
The standard axial design includes the following 2*q +m runs (q = number of mixture
factors, m centroid points as specified by user).
1. All the q vertex points. The coordinates of the ith Vertex point is
xi = (0, 0, 0..1, 0, 0..).
2. All q interior points of the simplex. The coordinates of the ith Interior point is
xi = (1/2q, 1/2q, 1/2q,..(q+1)/2q, 1/2q, 1/2q..).
3. The overall centroid of the simplex with coordinates
x = (1/q, 1/q,....., 1/q..) replicated (m-1) times.
265
User Guide to MODDE
RSM
MODDE provides 2 variants of the quadratic model designs, one special cubic and one
cubic. The simplex centroid design has all the experimental points on the vertices, and
on the center of the faces of consecutive dimensions.
Modified simplex centroid (SimM)
The modified simplex centroid design supports a quadratic model and includes the
following:
1. The q vertex points. The coordinates of the ith Vertex point is
xi = (0, 0, 0..1, 0, 0..).
2. The (q (q-1))/2 Edge centers. The coordinates of the ijth edge point is xij = (0,
0, 1/2, 1/2 0, 0..).
3. The q Interior check points. The coordinates of the ith interior point is xi =
(1/2q, 1/2q, (q+1)/2q, 1/2q, 1/2q..).
4. The overall centroid with coordinates x = (1/q, 1/q,...1/q), replicated as
desired.
266
Appendix B: Designs
267
User Guide to MODDE
D-Optimal designs
What are D-Optimal designs?
D-Optimal designs are computer generated designs, tailor made for a specific
problem. They allow great flexibility in the specifications of your problem. They are
particularly useful when you want to constrain the region and no classical design
exists.
“D-Optimal” means that these designs maximize the information in the selected set of
experimental runs with respect to a stated model.
For a specified regression model Y = X*β + ε where:
Y is a (N x 1) vector of observed responses,
X is a (N x p) extended design matrix, i.e. the n experimental runs extended with
additional columns to correspond to the p terms of the model (i.e., the added columns
are for the constant term, interaction terms, square terms, etc..)
β (beta) is a (p x 1) vector of unknown coefficients to be determined by fitting the
model to the observed responses.
ε (epsilon) is a (N x 1) vector of residuals (the differences between the observed and
predicted values of the response y). They are assumed to be independent of each other,
normally distributed and with constant variance σ2
The D-Optimal design maximizes the determinant of the X'X matrix, which is an
overall measure of the information in X. Geometrically; this corresponds to
maximizing the volume of X in a p dimensional space.
Candidate set
D-Optimal designs are constructed by selecting N runs from a candidate set. This
candidate set is the discrete set of “all potential good runs”.
MODDE generates the candidate set as follows:
I) For a regular process region, the candidate set consists of one or more of the
following sets of points (depending on your model and the number of factors):
• The full factorial for up to 10 factors, reduced factorial for up to 32 factors.
• Centers of edges between hyper-cube corners
• Centers of the faces of the hyper-cube.
• Overall centroid
II) For constrained regions of mixture or/and process factors, the candidate set consists
of one or more of the following set of points:
• The extreme vertices of the constrained region
• The centers of the edges. If these exceed 200, the center of the 200 longest
edges
• The centers of the various high dimensional faces
• The overall centroid.
MODDE has implemented an algorithm to compute the extreme vertices, center of
edges, center of faces etc. as described by Piepel (1988).
268
Appendix B: Designs
D-Optimal algorithm
D-Optimal designs have been criticized for being too dependent on an assumed model.
To reduce the dependence on an assumed model, MODDE has implemented a
Bayesian Modification of the K-Exchange algorithm of Johnson and Nachtsheim
(1983), as described by W. DuMouchel and B. Jones in “A Simple Bayesian
Modification of D-Optimal designs to reduce dependence on an Assumed Model”,
Technometrics (1994).
With this algorithm one can add to the “primary terms” i.e. the terms in the model,
“potential terms”, i.e. additional terms that might be important. The objective is to
select a D-Optimal design, rich enough to guard for potential terms, and enable the
analysis to detect possibly active ones.
In order not to increase the number of runs N, and to avoid a singular estimation, one
assumes that the coefficients of the potential terms are likely to have a mean of 0 and a
finite variance (tau, τ)2.
Potential terms
Potential terms are higher order terms not included in the model but taken into account
during the creation of the candidate set. Potential terms are default added but can be
removed by clearing the Use potential terms box.
Depending on the number of factors, the objective and the model, MODDE adds the
following potential terms:
RSM
Factors Model Potential terms
269
User Guide to MODDE
RSM
Factors Model Potential terms
7 - 12 Quadratic None
RSM
Factors Model Potential terms
Note: No potential terms are added for investigations with all factors defined
as qualitative.
270
Appendix B: Designs
Design evaluation
To evaluate and compare D-Optimal designs, MODDE computes the following
criteria:
LogDetNorm
The log of the determinant of X'X normalized for number of terms in the model p, and
number of runs N.
This is the criterion used, by default, to select the best design. MODDE selects the
design with the largest value (closest to 0) of LogDetNorm.
LogDetNorm = Log10 [ Det(X'X)1/p / N]
The maximum value of LogDetNorm, for an orthogonal design, is 0.
LogDet
The Log of the determinant of the X'X matrix
Condition No
The condition number of the X design matrix coded orthogonal, and extended
according to the model.
G efficiency
G efficiency is a lower bound on D efficiency, which compares the efficiency of a D-
Optimal design to a fractional factorial.
G efficiency is defined as:
Geff = (100*p)/(n*d)
Where
p = number of terms in the model
n = number of runs in the design
d = Maximum relative prediction variance v over the candidate set, where the
prediction variance v = x(X'X)-1x'
Note: All of these statistics are computed from the runs selected D-optimally
and do not include the possible center points added to the worksheet.
271
User Guide to MODDE
Irregular region
Screening
When the mixture region is an irregular polyhedron, MODDE computes the extreme
vertices (corners) delimiting the region. These extreme vertices constitute the candidate
set and the centers of the high dimensional faces are added to support potential
terms. The design is a D-optimal selection of N (specified by user) runs from the
candidate set.
RSM
MODDE computes the extreme vertices, 1/3, 2/3 centers of edges, centers of faces of
dimension (q-1) and the overall centroid of the experimental region. When there are
too many extreme vertices, only the center of the 25% longest edges is computed.
These experimental points constitute the candidate set.
The design is a D-Optimal selection of N runs (specified by the user) from the
Candidate set.
Mixture models
Because of the mixture constraint, (the mixture factors are not independent) the
analysis of mixture data with multiple regression requires a special model form.
The traditional approaches have been:
• Defining the model omitting one mixture factor, hence making the others
independent. This is the Slack Variable approach.
• Omitting some terms from the model, so that the terms remaining in the
model are independent. This is Scheffé model, with the constant term
removed from the linear model and the quadratic terms removed from the
quadratic model.
• Using the complete model including all the mixture terms, but putting
constraints on the coefficients to make them estimable. This is the Cox
reference model, and the constraints on the coefficients are defined with
respect to a standard reference mixture. This standard reference mixture
serves the same function as the centering constant with process variables
models.
272
Appendix B: Designs
If γ (gamma) are the coefficients of the interactions between the process and mixture
factors:
∑γksk = 0
Note: When the model contains terms of order 3, or contains qualitative and
formulation factors, the PLS coefficients are not adjusted relative to a stated
standard mixture.
273
User Guide to MODDE
274
Appendix C: Optimizer
Introduction
The optimizer works according to a given set of specifications. The specification of the
factors and responses are selected according to the desired result. A drawback with this
strategy can be that the best possible solution might not be reached. With unrealistic
response specifications it will be impossible for the optimizer to do a good job. With a
good strategy and by using complementary tools such as contour plots, DPMO
estimates, sweet spot plots and Design Space estimates, a good understanding for the
possibilities can be obtained.
The optimizer is used to find an experimental setpoint that fulfills various criteria. The
optimizer uses a search function to find the best possible solution to an equation
system given a number of operating criteria. The optimizer starts with a number of
criteria set in the optimizer window.
This appendix describes the possibilities and limitations of the optimizer function. The
first part is a description of how the optimizer works and the second part discusses how
different objectives can be reached by selecting different start criteria for the
optimization.
Search function
The optimizer works with a desirability function (f(ds)) that searches for the best
possible combination of factor settings that predicts a result inside the response
specifications and as close as possible to the target(s). When searching for one solution
with many criteria, the result will be a compromise. This compromise is based on a
summary function that is a measure of the distance to target for all results. It is
expressed as f(ds).
How f(ds) works depends on the weight for each response and the limit and target
specifications: the optimizer objective.
275
User Guide to MODDE
Optimizer objectives
The optimizer can be set up for different objectives:
1. Limit optimization – where the objective is to reach a solution where all
responses are within the specification limits. This is default in MODDE.
2. Target optimization – where the objective is to reach a solution where all
responses are as close to target as possible. For the target optimization it is
necessary that all responses can be optimized close to or to reach the target.
Otherwise you may end up with an unacceptable solution.
3. Focus optimization – where the objective is to favor one or several responses
over the others using individual weights.
To control the optimization criteria the weight function has a key role, as well as
reasonable limits and targets for the responses. The optimizer works with a desirability
function, f(ds), and will strive to reach the lowest possible value. The shape of the
function is controlled by the weight and the settings of criteria (Min, Target, Max) for
each response. In the following pictures two desirability functions are shown, the first
with weight=1 and the second with weight=0.2.
With the weight 1 the lowest possible f(ds) is -100 and with a weight of 0.2, the lowest
possible f(ds) is -20.
Limit optimization
With weight=1 the desirability function decreases rapidly close to the limit and then
flattens out. As consequence it will be easier to reach a compromise where all
responses are inside the specification limits but maybe not as close to target as
possible.
Weight = 1. The desirability function reached the lowest possible value, -100, just
inside the specification limit. This will work well to find a compromise when many
responses strive to get inside the specification limits. Note that in this plot the 'weight
scaled limit' and the 'limit' are displayed on top of each other.
276
Appendix C: Optimizer
Target optimization
If all weights are set to 0.2, the optimizer will search for a target solution for all
responses if that is possible. If that is not possible, the optimizer might find a solution
that will predict some responses very close to target and some outside the specification
limits.
Weight = 0.2. The desirability function will reach the lowest possible value, -20, close
to the target. If the weight is set to 0.2 for all responses it will go for a target solution
due to the slope of the function. This solution might be unacceptable if a common
target solution isn’t present.
Focus optimization
If the weights are set differently for different responses, responses with higher weights
take priority in the search for a solution inside the specifications. The overall
optimization criterion is to reach the lowest sum of f(ds). With the weight 1 the lowest
possible f(ds) is -100 and with a weight of 0.2, the lowest possible f(ds) is -20.
277
User Guide to MODDE
Replicate Plot: All values for NOx can be found between 10 and 35.
The response Soot is minimized while the response Fuel is set to Target for
illustrational purposes. Generally we also want to minimize the Fuel consumption.
278
Appendix C: Optimizer
Optimizer result
The optimizer will search for a solution to the specifications from 8 different starting
points using a Nelder-Mead Simplex algorithm trying to minimize the desirability
value f(ds). The result is expressed as a normalized distance to target Log(D) and
DPMO (Defects Per Million Opportunities outside specifications).
For more information about the calculations, see the Optimizer section in the
Statistical appendix.
After running the optimizer the best proposal (lowest Log(D)) is selected. A Log(D) <
-1 means that all results should be safely within specification limits. The optimal value
of Log(D) is -10; then all response predictions are on target. DPMO gives information
about robustness to small disturbances introduced by the Sensitivity Range specified
for the factors.
In the run list above, row 6 has a DPMO = 0 meaning that with disturbances + 5% on
the factor settings will give a solution inside the specifications. Row 1 has
approximately the same Log(D) but a DPMO = 7000 indicating that a small
disturbance in the factor settings from this point will result in some hits outside the
specifications. Therefore row 6 is preferable.
The result from the optimizer can be evaluated with three main tools:
• Contour Plot Wizard generates a plot around the factor settings of the
selected row, showing the dynamics around the selected point, e.g. if it is a
flat region or very sensitive to small changes. The selected run is displayed
in the plot as lines from the axes with arrows pointing toward the position of
the selected run.
• Analyze Design Space displays how the factor settings can be varied
around the selected point (optimum) and still fulfill the response criteria. We
can make an estimation of a safe DS region with Monte Carlo simulations on
the factor settings.
279
Appendix D: Design Space
Introduction
The calculation of the design space is a search function that expands the possible factor
ranges from a setpoint (optimum) to the largest possible range where all response
predictions are still within the specifications.
Predictions in the design space are done with Monte Carlo simulations. The resulting
distribution of predictions simulates a real situation with a random combination of
factor setting disturbances within a given range.
There are two Design Space features in MODDE:
• Predictive Design Space Estimation for optimization.
• Design Space Validation for robustness testing.
This chapter gives further insight to these features.
281
User Guide to MODDE
The predictive DS estimate shows the factor settings of the selected run in the
optimizer and the accepted variability around these settings that still result in
predictions in the accepted response region.
We can make an estimation of a safe DS region with Monte Carlo simulations on the
factor settings. MODDE will perform a search to identify the largest possible range for
each factor (faded region) that can be used and still meet all response requirements.
The default target is 1000 (0.1%) hits outside the limits (DPMO) for one response. In
this case the limiting response will be Soot, the final predictions will give 0.17% of the
predictions outside the upper limit.
282
Appendix D: Design Space
These settings give results within specifications but with a combination of factor
settings that is more extreme and close to the experimental limits. Another
consequence is that the accepted region of variability for the proposed factor settings is
narrower than the previous proposal starting from row 6.
283
User Guide to MODDE
284
Appendix D: Design Space
Evaluation
The Prediction | Design Space Validation tests the robustness by making a large
number of random disturbances (Monte Carlo Simulation) in the specified region. In
this example the specified region is the Experimental Region. In the DS window
(shown below) the factor part shows the original investigation settings with a specific
selection for the qualitative factor Column. ColA was manually selected as this column
gave the worst results (See Tutorial example for more information about the specific
example). The result is shown as a distribution of random samples including model
prediction errors and it is well within the specification limits for some responses. The
result can be expressed in general statistics as well as capability indexes Cpk or DPMO
= Defects Per Million Opportunities outside specifications.
A description of the details of this window is found in the Design Space window
section in the Design Space chapter.
Factor spreadsheet
All factors are varied within the design limits with Monte Carlo simulations according
to a Normal distribution. These are the default settings.
Response spreadsheet
The result for response k1 is optional; there are no specific demands for this response.
The result for response k2 is partially outside the specification limits.
The result for response Res1 is above the low specification limit.
The result for response PlateN(2) is above the low specification limit
From the above we conclude that this system is robust against disturbances in the
factors for Res1 and PlateN(2). k2 is not robust against disturbances in the factors.
285
User Guide to MODDE
Final adjustments
Design Space can be used to estimate the maximum accepted variability in factors that
still predict all results within the specifications.
The problem in the described example is response k2. The requirement for k2 is that
less than 0.1%, corresponding to DPMO = 1000, of the predictions may be outside the
specification limits.
There are constraints when handling this type of situation;
• Which factors affect the result?
• How can we adjust the factor limits without causing too much problems in
the normal use of the procedure?
First we have to check the model to understand which factors are the most influential.
The model has to be significant for an adjustment in factor ranges to have affect on the
result distribution.
In this example the model for k2 is very significant and the most important factor is
Acetonitrile (ACN).
286
Appendix D: Design Space
Assuming that the factor Temperature is easy to control with a narrower range we start
with this factor by adjusting temperature to +/- 0.5 °C. At the same time we can open
the Role for ACN to 'Free'. This instruction together with the specification limit for k2
(DPMO = 1000) will give an estimate of a range for ACN where we can predict that
the system is robust according to the specifications. The picture below displays the
result of the settings.
The proposed settings for ACN are now 25.51 to 26.49 and the estimated distribution
for k2 is 1410 hits outside the specification limits.
A final step might be to make an adjustment of the factor settings to some practical
new specification within the range for ACN, for instance 25.5 to 26.5. The result
shown below implies that the critical response k2 will have 0.16% of future predictions
outside the specifications.
287
User Guide to MODDE
To view statistics, right-click the Design Space window, and then click Design Space
Statistics.
A more detailed description of this example is found in the tutorial named “Robustness
Testing”.
288
References
289
User Guide to MODDE
290
Analysis
Index Fit............................................ 136, 137
Menu ......................................... 15, 133
Of variance ............................. 150, 238
Phase ............................................... 2, 9
Wizard ............................................ 135
Analysis Advisor ................................... 45
Analysis menu
2 Organization ................................... 133
Select Fit Method ............................ 136
2D contour ................................... 167, 187
Analysis of Variance ................... 150, 238
3 Analysis phase ......................................... 2
3D Analysis Wizard .................................. 135
Labels..............................................216
Onion plots......................................115 Arrow .................................................... 41
Rotation response surface .................41 Audit trail .............................................. 53
Scatter ............................. 115, 126, 176
Augmenting designs .............................. 23
Surface ............................................172
Zoom and rotate ................ 41, 214, 220 Auto update predictions....................... 176
Auto, MLR, PLS, Scheffé ................... 137
4
Automatic fit ......................................... 59
4D
Contour ...........................................169 Automatic update .......................... 59, 205
Sweet spot .......................................177 Autoscale modifier ................................ 77
Accelerators ......... 30, 31, 45, 64, 221, 224 AXR .................................................... 265
Activation ............................................229 B
Add Balanced ...................................... 106, 257
Command to Favorites ......................46
Block
Experiments ....................................131
Blocking............................................ 56
Factor .......................................... 27, 65
Blocks ....................................... 95, 102
Inclusions ..........................................90
BlockV ............................................ 131
Response ..................................... 32, 75
D-Optimal designs .......................... 252
Row ...................................................32
Interactions ............................. 102, 250
Squares and interactions.............. 34, 38
Mark ................................................. 41
Terms ................................................34
Orthogonal ...................................... 250
To Favorites ......................................51
Random..................................... 56, 253
To Report .................................. 51, 228
RSM designs ................................... 252
Add to Favorites ....................................51 Screening designs ........................... 251
Add to Report ........................................51 Box Behnken designs .......................... 260
Adjust according reference .......... 174, 243 Box Whisker........................................ 128
Alpha level ............................................54 Box-Cox plot ............................... 147, 243
291
User Guide to MODDE
292
Index
293
User Guide to MODDE
294
Index
Inclusions M
Added to the worksheet.....................90
Design augmentation ......................271 Main Effect.................................... 58, 157
Editing inclusions..............................92 Manage licenses .................................. 229
Generating.........................................92
Maximum runs .................................... 199
Part of the design ..............................91
vs. complement design ......................90 Menu item ............................................. 10
Increment.............................................207 Mid-range ...................... 70, 235, 236, 240
Insert Image .........................................226 Minimum ..................................... 207, 248
Insert Rows.............................. 32, 92, 176 Mip-file ................................................. 11
Installation ...............................................1 Missing ................................................ 237
Interactions .............................. 34, 99, 158 Mixture
and process factors .................. 247, 263
Interior points ......................................265
Constraint........................................ 264
Interpreting the model .........................152 Contour plot wizard ........................ 166
Investigation Options.............................52 Data................................................. 243
Designs ........................................... 263
Irregular region ....................................272 Experimental region ........................ 264
Factor definition.............................. 263
L Hierarchy ........................................ 243
Labels ..................................................176 Irregular region ............................... 272
Lack of Fit Models ............................................ 243
Alpha level ........................................54 Prediction plot wizard ..................... 173
ANOVA ..........................................238 MLR scaling .................... 70, 77, 235, 236
DF ...................................................134 MLR solution from PLS ...................... 234
Model validity .................................140
Plot ..................................................149 MODDE .................................... 1, 11, 230
Latent structures ..................................232 Model
Distance .......................................... 146
Layers .................... 59, 112, 258, 261, 274 Fit............................................ 136, 137
Layers overlap .....................................112 Hierarchy ........................................ 234
L-designs ................................... 5, 18, 258 Individual ........................................ 134
Interpret .......................................... 152
Legend ......................................... 210, 214 List .................................................. 204
Line style .............................................210 Saturated ......................................... 142
Select .......................................... 38, 95
Linked responses ...................................82
What is ................................................ 5
List presentation ....................................57
Model bar .............................................. 44
Loading plot ................................ 162, 163
Model predictive power....................... 233
Lock
Model validity ..................................... 140
Contour levels .................................213
Investigation......................................29 Multilevel factor .................................... 67
Log Multiple Linear Regression ............. 7, 231
Determinant ............................ 109, 271 Multiplots ...................................... 39, 206
In audit trail.......................................53
Multivariate ................................... 18, 232
LogDet.................................................271
LogdetNorm ................................ 109, 271 N
Low Limit............................................181 Network installation ............................ 229
New ......................................... 11, 17, 226
295
User Guide to MODDE
296
Index
Pseudo components ..................... 137, 200 Replicate tolerance ................ 59, 130, 201
Pure error ..................................... 134, 150 Report Generator ........................... 22, 221
Report Generator FAQ window .......... 226
Q Report/template ................................... 226
Q2 ................................................ 140, 239
Reproducibility ............................ 140, 141
Qualitative factors ................... 67, 82, 241
Reset button........................................... 34
Quantitative ............................. 67, 69, 263
Residual Standard Deviation ............... 237
Quantitative multilevel .............. 67, 68, 69
Residual types ..................................... 146
R Residuals ....................... 58, 144, 145, 241
R2 .................................. 57, 139, 142, 239 Residuals plots .................................... 144
Random ................................... 56, 61, 253 Resolution III ................................ 23, 257
Resolution IV designs ......................... 257
297
User Guide to MODDE
298
Index
Summary plot .............................. 139, 141 Unit Variance .......................... 70, 71, 235
Sweet Spot Plot ........................... 177, 189 Unit variance scaling ..................... 70, 235
299
User Guide to MODDE
300